Main
Date: 26 Jun 2008 09:28:25
From: Sanny
Subject: Zebediah made the King dance from e8-a1.
An interesting game by Zebediah against Advance Level..

Black King had to run out from e8 to a1. Zebediah trapped Black's Rook
and win easily.

Most interesting was how Black King ran from e8 to a1 But atlast has
to surrender.

Game Played between zebediah and advance at GetClub.com

--------------------------------------------------------------------------------
zebediah: (White)
advance: (Black)
Game Played at: http://www.GetClub.com/Chess.html
View Recorded Game: http://www.getclub.com/playgame.php?id=DM21199&game=Chess
--------------------------------------------------------------------------------

White -- Black
(zebediah) -- (advance)

1. d2-d4{16} d7-d5{0}
2. c2-c4{4} e7-e6{0}
3. Nb1-c3{4} Ng8-f6{0}
4. Ng1-f3{4} c7-c6{0}
5. e2-e3{4} d5-c4{1812}
6. Bf1-c4{17078} b7-b5{0}
7. Bc4-d3{8} Bf8-d6{1878}
8. e3-e4{712} b5-b4{1330}
9. e4-e5{834} b4-c3{1314}
10. b2-c3{58} Bc8-a6{3050}
11. e5-f6{976} g7-f6{2310}
12. Bd3-a6{310} Nb8-a6{3116}
13. Ke1-g1{3802} Na6-c7{1544}
14. Bc1-h6{810} Rh8-g8{1732}
15. Rf1-e1{866} Rg8-g6{1702}
16. Bh6-d2{396} Qd8-d7{1578}
17. Ra1-b1{192} c6-c5{2608}
18. Nf3-h4{790} Rg6-g8{1286}
19. Qd1-f3{24} Nc7-d5{1416}
20. c3-c4{142} Nd5-c7{1344}
21. Bd2-a5{864} Ra8-d8{2940}
22. Nh4-f5{11778} c5-d4{2592}
23. Nf5-d6{1240} Qd7-d6{1620}
24. Rb1-b7{298} Rd8-c8{1780}
25. Ba5-c7{118} Rc8-c7{2488}
26. Rb7-b8{7848} Ke8-e7{0}
27. Rb8-g8{4} Qd6-b4{1746}
28. Re1-e6{198} f7-e6{2934}
29. Rg8-g7{186} Ke7-d8{2282}
30. Qf3-a8{140} Rc7-c8{2056}
31. Rg7-g8{212} Kd8-e7{2946}
32. Qa8-a7{66} Ke7-d6{2432}
33. Qa7-d4{400} Kd6-c7{1936}
34. Rg8-g7{12} Kc7-c6{1310}
35. Qd4-d7{338} Kc6-c5{1916}
36. Qd7-c8{220} Kc5-d4{1990}
37. Qc8-d7{2912} Kd4-c3{1536}
38. Qd7-d1{288} Kc3-b2{1654}
39. Rg7-d7{66} Qb4-c4{1590}
40. Qd1-d2{1114} Kb2-a1{148}
--------------------------------------------------------------------------------
zebediah: (White)
advance: (Black)
Game Played at: http://www.GetClub.com/Chess.html
View Recorded Game: http://www.getclub.com/playgame.php?id=DM21199&game=Chess

What do you think Advance Level was wrong that it lost its Rook?

Can you spot any mistake in this interesting game?

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html




 
Date: 03 Jul 2008 01:54:02
From: Sanny
Subject: Re: When only 1 legal move, GetClub will not think more.
> Well, I'm glad your program is already playing good Chess.
>
> But learning a bit about how Chess programs work, such as learning
> about A-B pruning, would definitely help you improve your program
> still further, and it is not nearly as hard to do that as it would be
> to, say, add 400 points to your Elo.

After AB was used not much improvement was found. Only 30-40%
improvement found. Any other thing that can improve the program.

As Playing 30% faster keeps the game fast. So even if it has slow
improvement its worth it.

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html





  
Date: 03 Jul 2008 21:48:33
From: Guest
Subject: Re: When only 1 legal move, GetClub will not think more.
"Sanny" <[email protected] > wrote in message
news:11bedece-b0c2-4ab2-bb4b-1ca725ff0846@d19g2000prm.googlegroups.com...
>> Well, I'm glad your program is already playing good Chess.
>>
>> But learning a bit about how Chess programs work, such as learning
>> about A-B pruning, would definitely help you improve your program
>> still further, and it is not nearly as hard to do that as it would be
>> to, say, add 400 points to your Elo.
>
> After AB was used not much improvement was found. Only 30-40%
> improvement found. Any other thing that can improve the program.

That is a very interesting statement.

AlphaBeta (A-B) pruning should improve your search time by about a factor of
square root.

In other words, if it takes 16 minutes without it, then it will now take
only 4 minutes. If it takes 256 minutes, then it now only takes 16 minutes.
And so on.

If you are getting only 30$-40% improvement, then either you are doing it
wrong, or you might be using selective search.

Selective search does benefit from AB (and the other pruning methods) but
not as much as a plain brute force full width search.




>
> As Playing 30% faster keeps the game fast. So even if it has slow
> improvement its worth it.






----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
---= - Total Privacy via Encryption =---


 
Date: 01 Jul 2008 18:01:32
From: Quadibloc
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 1, 11:27 am, Sanny <[email protected] > wrote:

> Correct, Chess is very vast. You cannot learn all techniques. else
> every one will be a grand master.

Well, I'm glad your program is already playing good Chess.

But learning a bit about how Chess programs work, such as learning
about A-B pruning, would definitely help you improve your program
still further, and it is not nearly as hard to do that as it would be
to, say, add 400 points to your Elo.

John Savard


 
Date: 01 Jul 2008 17:18:10
From: help bot
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 1, 5:15 pm, Frisco Del Rosario <[email protected] > wrote:

> > I think the issue is you don't have a background in chess, or
> > programming. And depsite your efforts, I don't think you've gained
> > useful knowledge in either.

> That hit it right on the nose, but won't stop him, slow him down, or
> persuade him to learn anything.

With the amazing speed of todays computers,
perhaps a decent chess program can be made
without fancy pruners, hashing tables or other
chess-centric programming techniques. We
know that ignoring tactics leads to mediocrity,
but suppose someone were to focus on getting
the tactics right-- what then? In one of my
recent games, I landed a very shallow tactical
blow-- but suppose the speed and power of
modern computers were harnessed and aimed
squarely at this crucial aspect? Who knows
what would happen. I sometimes feel that my
computer's microprocessor is just spinning its
wheels with GetClub; that I'm out-running it
only because it is churning its own tires into
molten rubber while staying in the same spot
and burning lots of gas.


> Sanny and Ray Gordon ought to collaborate on a project.

Flash is the guy who keeps scolding
Sanny for supposedly spamming rgc, all the
while appending dozens of lines of real spam
to each of his own postings; this guy is
about as smart as a box of rocks. He
reminds me of nearly-IMnes in that respect.

But what Sanny really needs is a real, live
wolf; that way, when he screams "wolf!" and
people ask him "where?", he can just point
to his collaborator and say "right there".
Even a Siberian Husky would do.


-- help bot







  
Date: 02 Jul 2008 23:29:22
From: Patrick Volk
Subject: Re: When only 1 legal move, GetClub will not think more.
On Tue, 1 Jul 2008 17:18:10 -0700 (PDT), help bot
<[email protected] > wrote:

>On Jul 1, 5:15 pm, Frisco Del Rosario <[email protected]> wrote:
>
>> > I think the issue is you don't have a background in chess, or
>> > programming. And depsite your efforts, I don't think you've gained
>> > useful knowledge in either.
>
>> That hit it right on the nose, but won't stop him, slow him down, or
>> persuade him to learn anything.
>
> With the amazing speed of todays computers,
>perhaps a decent chess program can be made
>without fancy pruners, hashing tables or other
>chess-centric programming techniques. We
>know that ignoring tactics leads to mediocrity,
>but suppose someone were to focus on getting
>the tactics right-- what then? In one of my
>recent games, I landed a very shallow tactical
>blow-- but suppose the speed and power of
>modern computers were harnessed and aimed
>squarely at this crucial aspect? Who knows
>what would happen. I sometimes feel that my
>computer's microprocessor is just spinning its
>wheels with GetClub; that I'm out-running it
>only because it is churning its own tires into
>molten rubber while staying in the same spot
>and burning lots of gas.

When you write a chess program, I think it's a pretty fair statement
that you have to be knowledgeable in both fields, or at least know
someone.

Ignoring tactics also leads to worse than mediocrity in programming.
Hate to burst your bubble, but pruners and hash tables are hardly
chess-centric programming techniques. The A-B table is straight out of
game theory (which comes into more mundane things now such as search
engines), and the hash table is pretty much used whenever you use any
database (Apache uses them. Java has them as a class for pete's sake).
Why are they pretty standard in the chess game toolkit? Because
chess has a massive amount of possible moves. Get 5-6 ply and your
move set is at least in the billions. Every pieces adds at least an
order of magnitude. A queen for example can have up to 28 different
places it can move (a knight or king 8, a bishop/rook 14, and even a
pawn has 4 possible moves).
If you notice, every respectable chess program scores the position.
Why? Because that score is used for A-B pruning. Basically, you don't
trace down the less-than-favorible lines (basically, this determines
the breadth. Depth is difficulty). Sure, you can put your queen next
to your opponents' pawn to be taken in the next half ply, but if you
have better moves, do you really want to analyze the permutations of
that?
In other words, it's sorting. If you want to tell me a decent chess
program can be made without sorting, I would have to say it's possible
a roomful of monkeys can write Shakepeare. Possible, but EXTREMELY
unlikely.
The fact that such things didn't occur to Sanny very much implies
that his code is written as a journey. A journey that burns bridges.
It's one thing to write a chess program, and not know one of the
domains. It's either arrogant or ignorant to not know either.

>
>
>> Sanny and Ray Gordon ought to collaborate on a project.
>
> Flash is the guy who keeps scolding
>Sanny for supposedly spamming rgc, all the
>while appending dozens of lines of real spam
>to each of his own postings; this guy is
>about as smart as a box of rocks. He
>reminds me of nearly-IMnes in that respect.
>
> But what Sanny really needs is a real, live
>wolf; that way, when he screams "wolf!" and
>people ask him "where?", he can just point
>to his collaborator and say "right there".
>Even a Siberian Husky would do.
>

I'm not so sure Sanny would pass a Turing test at this point... If
nothing else, he's optimistic. His website about management and stuff
gave me a chuckle.
Kind of reminds me of my 20's, where a bunch of us would sit around,
and talk of starting up a company. But none of us really had an idea
on what we wanted to do. That's where a plan comes in.


>
> -- help bot
>
>
>
>


 
Date: 01 Jul 2008 10:27:18
From: Sanny
Subject: Re: When only 1 legal move, GetClub will not think more.
> Well, that answered my question about whether you used A-B pruning.
> The answer is no. Instead of making it a rule, you've made it an
> exception.
>
> I think the issue is you don't have a background in chess, or
> programming. And depsite your efforts, I don't think you've gained
> useful knowledge in either.

Correct, Chess is very vast. You cannot learn all techniques. else
every one will be a grand master.

I am just 1100 rated player. I just play Chess in my free time
Occasionaly Just for fun.

Even Beginner Level pushes my limits. So I do not understand how
others are winning against the higher levels.

I find only arround 10 / 1000 players at GetClub are able to win
higher levels Other 990 Players are beaten by even the Beginner Level.

A few of the Top players use other Commercial Program So Only 3
Players actually win with Brain Power.

Players who win with Brains and not taking other help are.

1. Help Bot.
2. Bonsai.
3. Chrisf.

Rest all are even weaker than the Beginner Level.

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html




  
Date: 02 Jul 2008 23:32:25
From: Patrick Volk
Subject: Re: When only 1 legal move, GetClub will not think more.
On Tue, 1 Jul 2008 10:27:18 -0700 (PDT), Sanny <[email protected] >
wrote:

>> Well, that answered my question about whether you used A-B pruning.
>> The answer is no. Instead of making it a rule, you've made it an
>> exception.
>>
>> I think the issue is you don't have a background in chess, or
>> programming. And depsite your efforts, I don't think you've gained
>> useful knowledge in either.
>
>Correct, Chess is very vast. You cannot learn all techniques. else
>every one will be a grand master.

That's not the issue. You need some basic strategies however.

>
>I am just 1100 rated player. I just play Chess in my free time
>Occasionaly Just for fun.

Me and my brothers play on occasion.

>
>Even Beginner Level pushes my limits. So I do not understand how
>others are winning against the higher levels.

THAT's the issue. Many things seem to be pushing the limits. You've
made the comment that the program is getting complicated.

>
>I find only arround 10 / 1000 players at GetClub are able to win
>higher levels Other 990 Players are beaten by even the Beginner Level.
>
>A few of the Top players use other Commercial Program So Only 3
>Players actually win with Brain Power.

Do you think you're attracting top-tier chess players with your
program?

>
>Players who win with Brains and not taking other help are.
>
>1. Help Bot.
>2. Bonsai.
>3. Chrisf.
>
>Rest all are even weaker than the Beginner Level.
>
>Bye
>Sanny
>
>Play Chess at: http://www.GetClub.com/Chess.html
>


 
Date: 29 Jun 2008 23:30:57
From: Sanny
Subject: When only 1 legal move, GetClub will not think more.
> =A0 Understood. =A0However, there is a titanic
> difference between "good versus bad" moves
> and evaluating what is or is not legal.
>
> =A0 If there are but *two* legal moves, both a
> human player and achessengine might
> crank away at them, evaluating deeply and
> considering which is better. =A0But when there
> is only one *legal* move, all the thinking and
> whiling away of hours in the world cannot
> ever change that fact. =A0To do so would

Yes, now I understand what you mean. You were not talking abt weak
moves but legal moves.

So this have been corrected now. If there is "only 1 legal move" it
will play that move in just 1 second instead of thinking for 3-5 min.

That will save a lot of players time when there is only one legal move
and he is playing with Master / Advance Level.

In a game such moves occur only 1/2/3 times. But that will save
GetClubs time in every game.

Play a few game and let me know, When there is (only one legal move )
if its plays legal moves in 1 second or not.

The programmer has corrected it. But not yet tested it. That you have
to find by playing a few games.

Play Chess at: http://www.GetClub.com/Chess.html

And by seeing recorded games we can see if it is playing such moves
instantaniously or not.

That was a very good advice.

Now in every game 1-2 min will be saved.

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html


  
Date: 04 Jul 2008 22:15:29
From: help bot
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 4, 1:41 pm, Patrick Volk <[email protected] > wrote:

> Good, you can look up acronyms.

This lack of perceptiveness is telling. Dr. Sir
Phillip IMnes would and indeed has pretended
to know funky stuff, to try and impress the
weak-minded. But I have no /need/ to do that
sort of thing; my record in the Big One we
fought in Grenada gives me -- and the entire
flight crew of my sub -- all the accolades I
could ever desire!

Seriously fella, if you are impressed by an
introductory class such as this one I took in
college a quarter-century ago, perhaps you
just need to get out more. Don't sweat it,
kid-- it may be obsolete knowledge. Here's
a bone, Fido: I don't know HTML (gasp!).


> And I'll reiterate, UNIX used hash tables. So did the 1ESS.

Who sold them? Was Bill Gates the only
prospective customer? What about the rest
of us-- the peons? (Zorro will school you, if
you refuse to respect us peons.)


> >> Any piece can have a minimum of zero moves. Oh, and I forgot the
> >> promotion.

> > That's why you'll never become world champ.
> >You remind me of my old pal, Mr. Bogolyubov... .

> Never aspired to be world champ.

That's what they all say, /after/ they realize
how pointless it would be to try. Unlike you,
I readily admit that I mulled it over a time or
two; but after I got scholars-mated a couple
of times by nerdy-looking kids who screamed
"CHECK!" so the whole world knew I got
crushed, I reconsidered. (Besides, how was
I supposed to get past Bobby Fischer, who
was expected to come back out of retirement
at any moment?)


> Again, do you have a point, other than contradiction? Anything
> constructive to add perhaps?

Yes. I would like to add that you have failed
to respond to your horrific blunder just below:


> >> A-B is selective search. Even the massively parallel solutions use
> >> some form of selective search, if for nothing else, to explore the
> >> best-looking lines first.

The Guest revealed that you were just lost
in space here-- what do you have to say for
your arrogant, snotty, wrong self?


> >> >> In other words, it's sorting. If you want to tell me a decent chess
> >> >> program can be made without sorting.

> > That was /your idea/, not mine.

> No, your idea. Your idea because you said it's possible to write a
> program without using existing strategies.

Wrong, again. Do I begin to detect a
/pattern/ here?


> > Anyway, your suggestion to speed up the search
> >was a good one, except that it is even more crucial
> >for him to fix the problem with quiessence-search
> >first.

> WTF is quiessence?

It is the way we misspell the fancy word for "quiet
position" where I come from (as I already indicated).
Are you always this slow?


> Do you mean quiescence?

You tell me, Mr. Know-it-all No-help.


> I still contend fix the strategy

Shift duly noted. Your original position was
that Sanny needed to increase *speed*.


> If you have a barge cobbled together with
> garbage, it ain't going to be graceful, no matter how much chrome you
> put on it.

Agreed. I think the results indicate that
no matter how fast or slow, the tactics are
being seriously mishandled, and my guess
is that his programmers don't know about
the importance of "quiet positions" and
their relationship to calculating tactics in
chess. I have consistently pounded the
table, screaming at the top of my lungs
that I can not and will not be beaten by a
player/program which simply hangs its
pieces to me! (Well, not often, anyway.)

I reject the idea -- regardless of source --
that merely speeding things up will solve
this problem, because even at the lower
levels it should not exist-- yet it does.

To simplify, calculating tactics correctly
is more important than strategy, and this
ought to be priority number one (apart from
playing chess /in accordance with the
rules/).

Got anything to help Sanny on this? Like
say, a link to a Web site which explains
precisely how it is done. Or are you just
going to continue your jabbering?

I have (repeatedly) admitted not knowing
enough about this to explain how it is done.
You obviously also know very little (but you
might know someone, who knows someone,
who knows how it is done). ; >D


-- help bot



   
Date: 05 Jul 2008 18:22:27
From: Patrick Volk
Subject: Re: When only 1 legal move, GetClub will not think more.
On Fri, 4 Jul 2008 22:15:29 -0700 (PDT), help bot
<[email protected] > wrote:

>On Jul 4, 1:41 pm, Patrick Volk <[email protected]> wrote:
>
> Seriously fella, if you are impressed by an
>introductory class such as this one I took in
>college a quarter-century ago, perhaps you
>just need to get out more. Don't sweat it,
>kid-- it may be obsolete knowledge. Here's
>a bone, Fido: I don't know HTML (gasp!).
>

I'm a systems programmer, and don't know HTML either. There's programs
to do tha anyway.

>
>> And I'll reiterate, UNIX used hash tables. So did the 1ESS.
>
> Who sold them? Was Bill Gates the only
>prospective customer? What about the rest
>of us-- the peons? (Zorro will school you, if
>you refuse to respect us peons.)
>

Little company around in the 80's called Bell Telephone. Maybe heard
of them.

>
>> >> Any piece can have a minimum of zero moves. Oh, and I forgot the
>> >> promotion.
>
>> > That's why you'll never become world champ.
>> >You remind me of my old pal, Mr. Bogolyubov... .
>
>> Never aspired to be world champ.
>
> That's what they all say, /after/ they realize
>how pointless it would be to try. Unlike you,
>I readily admit that I mulled it over a time or
>two; but after I got scholars-mated a couple
>of times by nerdy-looking kids who screamed
>"CHECK!" so the whole world knew I got
>crushed, I reconsidered. (Besides, how was
>I supposed to get past Bobby Fischer, who
>was expected to come back out of retirement
>at any moment?)

Getting regular board beat-downs from my brothers made me find other
avenues. Funny, I taught my cousin to play chess. I was 9, he was 10.
He was beating me 5 minutes later. I think I'm not good stategically.

>
>
>> Again, do you have a point, other than contradiction? Anything
>> constructive to add perhaps?
>
> Yes. I would like to add that you have failed
>to respond to your horrific blunder just below:
>
>
>> >> A-B is selective search. Even the massively parallel solutions use
>> >> some form of selective search, if for nothing else, to explore the
>> >> best-looking lines first.
>
> The Guest revealed that you were just lost
>in space here-- what do you have to say for
>your arrogant, snotty, wrong self?

I shorted searching for moves to search. Mea cupla already.

>
>
>> >> >> In other words, it's sorting. If you want to tell me a decent chess
>> >> >> program can be made without sorting.
>
>> > That was /your idea/, not mine.
>
>> No, your idea. Your idea because you said it's possible to write a
>> program without using existing strategies.
>
> Wrong, again. Do I begin to detect a
>/pattern/ here?

You didn't say:

With the amazing speed of todays computers,
perhaps a decent chess program can be made
without fancy pruners, hashing tables or other
chess-centric programming techniques.

I countered they're not chess-centric techniques (A-B pruning is game
theory 101, and hashing is data structure 101). Hash tables are so
chess centric, they've been in just about everything that's come down
the programming pike in the last 15-20 years.

>
>
>> > Anyway, your suggestion to speed up the search
>> >was a good one, except that it is even more crucial
>> >for him to fix the problem with quiessence-search
>> >first.
>
>> WTF is quiessence?
>
> It is the way we misspell the fancy word for "quiet
>position" where I come from (as I already indicated).
>Are you always this slow?
>
>
>> Do you mean quiescence?
>
> You tell me, Mr. Know-it-all No-help.
>
>
>> I still contend fix the strategy
>
> Shift duly noted. Your original position was
>that Sanny needed to increase *speed*.

No... I said he needs to be more efficient, and it will result in
speed. Probably a smaller memory footprint.

>
>
>> If you have a barge cobbled together with
>> garbage, it ain't going to be graceful, no matter how much chrome you
>> put on it.
>
> Agreed. I think the results indicate that
>no matter how fast or slow, the tactics are
>being seriously mishandled, and my guess
>is that his programmers don't know about
>the importance of "quiet positions" and
>their relationship to calculating tactics in
>chess. I have consistently pounded the
>table, screaming at the top of my lungs
>that I can not and will not be beaten by a
>player/program which simply hangs its
>pieces to me! (Well, not often, anyway.)

I pretty much agree, but as a programmer.

My contention is he needs to go to the root of the problem. Scoring is
that. You can score the pieces you have, and the squares you control.
He needs to start there. Not hard.

Once you know where you are, then you can look ahead. If you spend
most of your allotted time seeing how you can recover after you hang
your queen out to dry, it's not hard to figure out your program would
suck.

>
> I reject the idea -- regardless of source --
>that merely speeding things up will solve
>this problem, because even at the lower
>levels it should not exist-- yet it does.

A-B pruning gives you time. That doesn't necessarily mean faster. That
means more time to do more productive things.

From what I've seen, and what people have said, including yourself, it
craps the bed every so often. It will play ok for a couple of moves,
and then do a really hideous one.
What I'm trying to point out is it appears that it gets a bunch of
bad lines, and starts exploring them. If there's enough of them,
you're only going to come up with bad moves. Better in that case to go
wide, look at more moves until you find one that is more promising
instead of looking at the first one, and going a ply or two in. I
didn't think I was being that vague.

>
> To simplify, calculating tactics correctly
>is more important than strategy, and this
>ought to be priority number one (apart from
>playing chess /in accordance with the
>rules/).

Agreed. To simplify it more, he needs to identify the best move and
make it. Opening books are for strategy. Endgame bases are strategy.

>
> Got anything to help Sanny on this? Like
>say, a link to a Web site which explains
>precisely how it is done. Or are you just
>going to continue your jabbering?

If he's expecting to make money off of this, he can do his own
research. It's bad enough he can't do any testing on his own so it
seems. That's where I draw the line. He or his programmer(s) need to
read a few books. Systems Programming by Silberschlatz, and any of the
game programming books out there will give you info.

>
> I have (repeatedly) admitted not knowing
>enough about this to explain how it is done.
>You obviously also know very little (but you
>might know someone, who knows someone,
>who knows how it is done). ;>D

Never said I know how to write a chess program, but I know something
about game theory.

I've worked in programming for a while now, and have done a bit,
mostly in the area of safety. I do telecommunications now.
And that is why I'm saying what I am. Even if you have a hat that
says "Lion Tamer" on it, and want to be a lion tamer, it's no
predictor of future performance as much as being a librarian. Wanna
write a chess program, you should either know chess or programming.



>
>
> -- help bot


  
Date: 01 Jul 2008 11:37:43
From: Patrick Volk
Subject: Re: When only 1 legal move, GetClub will not think more.
On Sun, 29 Jun 2008 23:30:57 -0700 (PDT), Sanny
<[email protected] > wrote:

>> � Understood. �However, there is a titanic
>> difference between "good versus bad" moves
>> and evaluating what is or is not legal.
>>
>> � If there are but *two* legal moves, both a
>> human player and achessengine might
>> crank away at them, evaluating deeply and
>> considering which is better. �But when there
>> is only one *legal* move, all the thinking and
>> whiling away of hours in the world cannot
>> ever change that fact. �To do so would
>
>Yes, now I understand what you mean. You were not talking abt weak
>moves but legal moves.
>
>So this have been corrected now. If there is "only 1 legal move" it
>will play that move in just 1 second instead of thinking for 3-5 min.
>
>That will save a lot of players time when there is only one legal move
>and he is playing with Master / Advance Level.
>
>In a game such moves occur only 1/2/3 times. But that will save
>GetClubs time in every game.
>
>Play a few game and let me know, When there is (only one legal move )
>if its plays legal moves in 1 second or not.
>
>The programmer has corrected it. But not yet tested it. That you have
>to find by playing a few games.
>
>Play Chess at: http://www.GetClub.com/Chess.html
>
>And by seeing recorded games we can see if it is playing such moves
>instantaniously or not.
>
>That was a very good advice.
>
>Now in every game 1-2 min will be saved.
>
>Bye
>Sanny


Well, that answered my question about whether you used A-B pruning.
The answer is no. Instead of making it a rule, you've made it an
exception.

I think the issue is you don't have a background in chess, or
programming. And depsite your efforts, I don't think you've gained
useful knowledge in either.



>
>Play Chess at: http://www.GetClub.com/Chess.html


   
Date: 01 Jul 2008 14:15:11
From: Frisco Del Rosario
Subject: Re: When only 1 legal move, GetClub will not think more.
In article <[email protected] >,
Patrick Volk <[email protected] > wrote:

> I think the issue is you don't have a background in chess, or
> programming. And depsite your efforts, I don't think you've gained
> useful knowledge in either.

That hit it right on the nose, but won't stop him, slow him down, or
persuade him to learn anything.

Sanny and Ray Gordon ought to collaborate on a project.


    
Date: 05 Jul 2008 22:05:33
From: Sanny
Subject: Re: When only 1 legal move, GetClub will not think more.
> > >> I still contend fix the strategy
>
> > > =A0Shift duly noted. =A0Your original position was
> > >that Sanny needed to increase *speed*.
>
> > No... I said he needs to be more efficient, and it will result in
> > speed. Probably a smaller memory footprint.
>
> =A0 Okay, more efficient is good. =A0But hisGetClub
> program has some fundamental problem which
> goes deeper than just slowness; it in fact
> miscalculates tactics, or perhaps it doesn't
> bother to try to calculate them-- as if that will
> help.

Have you played a game recently? Last 4 days the game wasagain
improved with lots of new things.

Play Chess at: http://www.GetClub.com/Chess.html

Have a game and tell me how you face it.

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html


    
Date: 05 Jul 2008 20:08:33
From: help bot
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 5, 6:22 pm, Patrick Volk <[email protected] > wrote:

> Little company around in the 80's called Bell Telephone. Maybe heard
> of them.

I was talking about /who sold their chess
machines/programs/ (i.e. the chess connection).


> > The Guest revealed that you were just lost
> >in space here-- what do you have to say for
> >your arrogant, snotty, wrong self?

> I shorted searching for moves to search.

Whatever that means. (Lots of people
here have trouble admitting they have no
clue what they rant about).


> > Wrong, again. Do I begin to detect a
> >/pattern/ here?
>
> You didn't say:
>
> With the amazing speed of todays computers,
> perhaps a decent chess program can be made
> without fancy pruners, hashing tables or other
> chess-centric programming techniques.
>
> I countered they're not chess-centric techniques (A-B pruning is game
> theory 101

You're splitting hairs; game theory/chess
game theory are the same thing, to me (i.e.
chess is a game).

I could have put in a hyphen like this, to
clarify my meaning: without fancy-pruners,
or maybe italics, like so: without /fancy/
pruners, since I was not discussing the
most elementary type, which merely cuts
out the obvious chunks of fat.


> and hashing is data structure 101). Hash tables are so
> chess centric, they've been in just about everything that's come down
> the programming pike in the last 15-20 years.

They weren't used by the earliest chess
programmers who actually sold their wares
to the public. The "experts" used to write
about how it was impractical to implement
hash-tables, due to limitations of memory,
etc.


> >> I still contend fix the strategy
>
> > Shift duly noted. Your original position was
> >that Sanny needed to increase *speed*.
>
> No... I said he needs to be more efficient, and it will result in
> speed. Probably a smaller memory footprint.

Okay, more efficient is good. But his GetClub
program has some fundamental problem which
goes deeper than just slowness; it in fact
miscalculates tactics, or perhaps it doesn't
bother to try to calculate them-- as if that will
help.


> I pretty much agree, but as a programmer.

There are worse things; lawyers, politicians,
slime, scum, fungi, lice-- but I keep repeating
myself.


> My contention is he needs to go to the root of the problem. Scoring is
> that. You can score the pieces you have, and the squares you control.
> He needs to start there. Not hard.

There are in fact some scoring issues-- such
as keeping its King near the corner in the
endgame. But this is not a big issue if the
program could only trounce humans at tactics,
before the endgame.


> Once you know where you are, then you can look ahead. If you spend
> most of your allotted time seeing how you can recover after you hang
> your queen out to dry, it's not hard to figure out your program would
> suck.

I myself have adopted a strategy in which
I allocate my OTB thinking time toward the
early part of the game, hoping to obtain a
winning position instead of a losing one.
Sanny's program does not compete in such
events; he decides the thinking times, along
with everything else (sometimes including
who wins). But people would (once again)
begin to complain, if he were to allocate a
lot more time.


> > I reject the idea -- regardless of source --
> >that merely speeding things up will solve
> >this problem, because even at the lower
> >levels it should not exist-- yet it does.

> A-B pruning gives you time. That doesn't necessarily mean faster. That
> means more time to do more productive things.

True. But he still has to write the code
so that the program /will/ do those things.


> From what I've seen, and what people have said, including yourself, it
> craps the bed every so often. It will play ok for a couple of moves,
> and then do a really hideous one.
> What I'm trying to point out is it appears that it gets a bunch of
> bad lines, and starts exploring them. If there's enough of them,
> you're only going to come up with bad moves. Better in that case to go
> wide, look at more moves until you find one that is more promising
> instead of looking at the first one, and going a ply or two in. I
> didn't think I was being that vague.

It seems to me that you must never have
played the program; how else could you
believe that widening the search will have
any great impact? You do not seem to
understand just how messed up are the
program's calculation of /basic tactics/.


> > To simplify, calculating tactics correctly
> >is more important than strategy, and this
> >ought to be priority number one (apart from
> >playing chess /in accordance with the
> >rules/).


> > Got anything to help Sanny on this? Like
> >say, a link to a Web site which explains
> >precisely how it is done. Or are you just
> >going to continue your jabbering?

> If he's expecting to make money off of this, he can do his own
> research. It's bad enough he can't do any testing on his own so it
> seems. That's where I draw the line. He or his programmer(s) need to
> read a few books. Systems Programming by Silberschlatz, and any of the
> game programming books out there will give you info.

I suppose that "answers" my question above.
I think Sanny is hiring professional programmers
to write the code; if so, your suggestion-- for
them to read books -- is not going to be well
received, nor will it likely help.


> I've worked in programming for a while now, and have done a bit,
> mostly in the area of safety. I do telecommunications now.
> And that is why I'm saying what I am. Even if you have a hat that
> says "Lion Tamer" on it, and want to be a lion tamer, it's no
> predictor of future performance as much as being a librarian. Wanna
> write a chess program, you should either know chess or programming.

A look at Sanny's Web site reveals that his
programmers know how to write code. But
where they fall down is in knowing chess, and
in all those subtle areas of programming
which relate specifically to speed and
efficiency, which can yield very strong chess
moves; such speed and efficiency as is not
normally required for many business
programs, on modern hardware. These things
are what many chess programmers obsess
over, because they have a /need/ for speed.


-- help bot



     
Date: 06 Jul 2008 23:29:35
From: Patrick Volk
Subject: Re: When only 1 legal move, GetClub will not think more.
On Sat, 5 Jul 2008 20:08:33 -0700 (PDT), help bot
<[email protected] > wrote:

>On Jul 5, 6:22 pm, Patrick Volk <[email protected]> wrote:
>
>> Little company around in the 80's called Bell Telephone. Maybe heard
>> of them.
>
> I was talking about /who sold their chess
>machines/programs/ (i.e. the chess connection).
>

I'm splitting hairs?

Bell might have used some of the lessons learned from chess to go into
the switching fabric. If they did that, they profited from it.

>
>> > The Guest revealed that you were just lost
>> >in space here-- what do you have to say for
>> >your arrogant, snotty, wrong self?
>
>> I shorted searching for moves to search.
>
> Whatever that means. (Lots of people
>here have trouble admitting they have no
>clue what they rant about).

Indeed.

>
>
>> > Wrong, again. Do I begin to detect a
>> >/pattern/ here?
>>
>> You didn't say:
>>
>> With the amazing speed of todays computers,
>> perhaps a decent chess program can be made
>> without fancy pruners, hashing tables or other
>> chess-centric programming techniques.
>>
>> I countered they're not chess-centric techniques (A-B pruning is game
>> theory 101
>
> You're splitting hairs; game theory/chess
>game theory are the same thing, to me (i.e.
>chess is a game).

Game theory can be used in the routing of trains, or how to place
your whiskey casks in a warehouse to move them the least (and store
the most).

>
> I could have put in a hyphen like this, to
>clarify my meaning: without fancy-pruners,
>or maybe italics, like so: without /fancy/
>pruners, since I was not discussing the
>most elementary type, which merely cuts
>out the obvious chunks of fat.

I'm bad for arguing sematics however ;)

>
>
>> and hashing is data structure 101). Hash tables are so
>> chess centric, they've been in just about everything that's come down
>> the programming pike in the last 15-20 years.
>
> They weren't used by the earliest chess
>programmers who actually sold their wares
>to the public. The "experts" used to write
>about how it was impractical to implement
>hash-tables, due to limitations of memory,
>etc.
>
>
>> >> I still contend fix the strategy
>>
>> > Shift duly noted. Your original position was
>> >that Sanny needed to increase *speed*.
>>
>> No... I said he needs to be more efficient, and it will result in
>> speed. Probably a smaller memory footprint.
>
> Okay, more efficient is good. But his GetClub
>program has some fundamental problem which
>goes deeper than just slowness; it in fact
>miscalculates tactics, or perhaps it doesn't
>bother to try to calculate them-- as if that will
>help.
>
>
>> I pretty much agree, but as a programmer.
>
> There are worse things; lawyers, politicians,
>slime, scum, fungi, lice-- but I keep repeating
>myself.

Hey, at least I have a job, unlike a few chess players...

>
>
>> My contention is he needs to go to the root of the problem. Scoring is
>> that. You can score the pieces you have, and the squares you control.
>> He needs to start there. Not hard.
>
> There are in fact some scoring issues-- such
>as keeping its King near the corner in the
>endgame. But this is not a big issue if the
>program could only trounce humans at tactics,
>before the endgame.

If you're writing a chess program, and you give bonus points for the
position if the king is not in a corner.

>
>
>> Once you know where you are, then you can look ahead. If you spend
>> most of your allotted time seeing how you can recover after you hang
>> your queen out to dry, it's not hard to figure out your program would
>> suck.
>
> I myself have adopted a strategy in which
>I allocate my OTB thinking time toward the
>early part of the game, hoping to obtain a
>winning position instead of a losing one.
>Sanny's program does not compete in such
>events; he decides the thinking times, along
>with everything else (sometimes including
>who wins). But people would (once again)
>begin to complain, if he were to allocate a
>lot more time.

Exactly why you need to be efficient. But if you're expecting a minute
a move, and the program decides it needs to take 5, what do you expect
the person to do?

>
>
>> > I reject the idea -- regardless of source --
>> >that merely speeding things up will solve
>> >this problem, because even at the lower
>> >levels it should not exist-- yet it does.
>
>> A-B pruning gives you time. That doesn't necessarily mean faster. That
>> means more time to do more productive things.
>
> True. But he still has to write the code
>so that the program /will/ do those things.

Do you think he does analysis? He got Rybka, but all he does is have
it play GC every few days, and posts the result here. He doesn't
analyze the games it plays. He /doesn't even know what's a legal
move/.


>
>
>> From what I've seen, and what people have said, including yourself, it
>> craps the bed every so often. It will play ok for a couple of moves,
>> and then do a really hideous one.
>> What I'm trying to point out is it appears that it gets a bunch of
>> bad lines, and starts exploring them. If there's enough of them,
>> you're only going to come up with bad moves. Better in that case to go
>> wide, look at more moves until you find one that is more promising
>> instead of looking at the first one, and going a ply or two in. I
>> didn't think I was being that vague.
>
> It seems to me that you must never have
>played the program; how else could you
>believe that widening the search will have
>any great impact? You do not seem to
>understand just how messed up are the
>program's calculation of /basic tactics/.

You're talking about a chess program in the manner of chess, and I'm
trying to put a programming spin on it. Basically a chess program:

1) Determines its' current position
2) Looks at the possible moves (and counter-moves, several moves ahead
if possible)
3) And chooses the best move

Tactics is (3). (2) and (3) need to happen concurrently.

I've played it twice, and have better things to do than beta test.
I'll play people on Pogo... better time controls.

>
>
>> > To simplify, calculating tactics correctly
>> >is more important than strategy, and this
>> >ought to be priority number one (apart from
>> >playing chess /in accordance with the
>> >rules/).
>
>
>> > Got anything to help Sanny on this? Like
>> >say, a link to a Web site which explains
>> >precisely how it is done. Or are you just
>> >going to continue your jabbering?
>
>> If he's expecting to make money off of this, he can do his own
>> research. It's bad enough he can't do any testing on his own so it
>> seems. That's where I draw the line. He or his programmer(s) need to
>> read a few books. Systems Programming by Silberschlatz, and any of the
>> game programming books out there will give you info.
>
> I suppose that "answers" my question above.
>I think Sanny is hiring professional programmers
>to write the code; if so, your suggestion-- for
>them to read books -- is not going to be well
>received, nor will it likely help.

Do I care, really? They can get offended by my comment, or they can
do something about it. I'd prefer if they did the latter. I don't
think it's the work of professional programmers.
When I work in a particular field, it's in my best interest to
figure out what they do, and what the rules are. I don't mind reading
to get the job done... Otherwise I'd be harping about how I took
FORTRAN IV in college, like some people.

>
>
>> I've worked in programming for a while now, and have done a bit,
>> mostly in the area of safety. I do telecommunications now.
>> And that is why I'm saying what I am. Even if you have a hat that
>> says "Lion Tamer" on it, and want to be a lion tamer, it's no
>> predictor of future performance as much as being a librarian. Wanna
>> write a chess program, you should either know chess or programming.
>
> A look at Sanny's Web site reveals that his
>programmers know how to write code.

Not good code.

> But
>where they fall down is in knowing chess, and
>in all those subtle areas of programming
>which relate specifically to speed and
>efficiency, which can yield very strong chess
>moves; such speed and efficiency as is not
>normally required for many business
>programs, on modern hardware.

Efficiency is MUCH more imporant than speed. Neither really is
important in business.

If you're doing pure analysis, then you're probably going to want
speed.

>These things
>are what many chess programmers obsess
>over, because they have a /need/ for speed.

Are you a chess programmer?

I can tell you even if you have the fastest machine in the world, a
crappy program will still bring it to its' knees. More importantly is
what you do with it. I bet the Fidelity box would trounce GC.


>
>
> -- help bot


 
Date: 29 Jun 2008 22:22:36
From: help bot
Subject: Re: 4 new Openings taught.

Sanny wrote:

> > > If you find any opening move missing let me know it.GetClubcan
> > > remember first four moves. So it plays opening moves in 0 seconds
> > > saving your time.

> > About that: when there is only one legal
> > move the program goes into a deep think,
> > as usual. But apart from figuring out that
> > there is one and only one *legal* move,
> > there's really nothing to think about, is
> > there? Just play the move on the board,
> > and crank away on the opponent's time.

> GetClub thinks when there is 1 forced move So that may be by thinking
> higher depth it may give a Gambit of Rook/ Queen to win the game?

If such a follow-up exists, it will still be there
after the (forced) only-legal-move is played and
the opponent begins thinking.


> Or may be some move may be found which draws the game by repetitive
> moves.

This could pose a technical problem.


> As many times some good move may look wrong at lower depth but only
> when depth of search is increased we can see it's benefits.

After the move is executed on the board, the
program ought to be able to "see" a bit deeper
into the subsequent position.


> I see in many games Zrbediah gives away its bishop/ Rook to win the
> game So for that reasion all moves must be evaluated. Even if they are
> giving loss at lower depth.

Understood. However, there is a titanic
difference between "good versus bad" moves
and evaluating what is or is not legal.

If there are but *two* legal moves, both a
human player and a chess engine might
crank away at them, evaluating deeply and
considering which is better. But when there
is only one *legal* move, all the thinking and
whiling away of hours in the world cannot
ever change that fact. To do so would
require changing the rules of the game-- an
act best reserved for execution in between
matches or tourneys.

A simple way to handle this is to churn
out a list of all possible moves, then chop
off the illegal ones, thus creating a list of
only the *legal* moves. Now, before the
next step (whatever it may be), you can
look to see if the list of *legal* moves you
just created has only one element, and if
so, designate it as the "best" move (you
can't be wrong about this) and play it on
the board (or check to see if a draw
condition exists, etc.).

------------------------------------------------------------

Suppose you have made a million more
of your famous "improvements", and are
now ready for a $100,000 match against
team Rybka. You could "pocket" the
thinking time saved here and use it on the
next move-- a move where you actually
have a choice of legal moves. In most
cases, the time controls are not such
that moving immediately like this just
throws away time you otherwise could
have utilized constructively. In real
tournaments and matches, the time
limits allot a chunk of time, to be used
as the program sees fit over the course
of a set number of moves. Under such
real-world conditions, your policy of
utilizing the full time limit on a *forced*
move would be a handicap. If I were the
computer operator for GetClub -- entitled
to fully half the purse if we won -- and
this situation occurred (after say, Q-d8+),
I would go ahead and play the only-legal-
move (...Kxd8) immediately and then sit
back and watch Rybka spit out her reply
(Bg5 double-check). Then I would try
asking for a delay or rescheduling of
(...Ke8, forced) some sort, and if that did
not work, I would be forced to use Plan B,
in which I activate a powerful magnet to
avoid having to face (Rd8++) reality. The
magnet would destroy both computers,
and every cellphone and then I would set
off a hidden bomb, destroying all
evidence that we had lost or even been
stupid enough to play such a match.

Anyway, it was just an idea.


-- helpful bot


  
Date: 11 Jul 2008 18:40:19
From: help bot
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 10, 10:54 am, "Guest" <[email protected] > wrote:

> >> Also, I'm a little suspicious that the program speed is doubling every
> >> month. That means in 5 months it's 32 times faster than before. This is
> >> *not*, repeat *NOT*, the same as comparing the play to Rybka or anything
> >> like that, but a real, genuine faster program (like if you moved to a
> >> computer that was 32 times faster than it was 5 months ago.)

> > Note that Sanny's program started out
> > well below the 1000 USCF level, based on
> > casual observation. So if he were to add
> > a whopping 600 points due to repeated
> > speed-ups, this is not so far-fetched as
> > you might imagine.

> I'm not so sure he's actually talking about doubling of his program's speed.
>
> He seems to be talking more about his program finding the claimed 'right'
> rybka answer that is increasing.


In reality, he has done *both*.


> That's not quite the same thing. Which is why I made sure I stressed that.
>
> Those kind of improvements are the 'fuzzy' kind that require formal testing
> methods to estimate ratings. You can't estimate based on the Tech curve.
>
> For his program to start below 1000 rating is pretty significant... How the
> EXPLETIVE can you get a proram to play that poorly on today's hardware??!
> Even a program that looks at only material and mobility should be stronger
> than that.


Precisely. That other guy -- the one who
started claiming that all the "experts" agree
with him, ala Nick Bourbaki -- can't seem to
see the forest for the trees.

My suggestion to Sanny was to stop his
"improvement" nonsense-talk and try to
home-in on a correct handling of tactics.

The multitude of tied position scores can
be broken by factors such as mobility--
which is an example of what made the
now-ancient Fidelity tabletop computers
so good (relatively speaking). Those old-
timers got the basic tactics right-- and
that included the Q-search, and not
deliberately sacrificing pieces for only
two pawns and a spite-check as Sanny's
program often does.

Personally, I think that it might also aid
in reducing the number of alleged bugs he
keeps inventing as scapegoats for the
multitude of defeats the program suffers.
Fewer defeats equal a reduced need for
bogus excuses. And how do you get
fewer defeats? By improving at tactics.


-- help bot



  
Date: 11 Jul 2008 14:33:21
From:
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 9, 2:57=A0pm, "[email protected]" <[email protected] > wrote:
> > MyChess did have two problems though.,,, (etc.)
>
> Very interesting info about this classic chess program. =A0However,
> consider this: my compressed DOS binary is about 20k in size, and that
> includes CGA graphics! =A0The program does a lot, considering its age
> and its incredibly tiny footprint. =A0I have much admiration for what
> the programmers of that era were able to do with relatively limited
> computing resources. =A0In the end MyChess plays a decent enough game
> for the most part, and even plays by the rules! =A0Sanny might learn
> from this.

For those who are not aware, Kittinger updated his engine for the
"Majestic Chess" program that was released about five years ago. The
engine played pretty well, although the complex interface and graphics
requirements of the game (it was designed to be attractive and give
competition to Chessmaster) meant that the engine itself could only
use about 50-70% of the CPU. Not sure what will happen on dual-core
systems now.

jm


  
Date: 10 Jul 2008 17:30:44
From: help bot
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 10, 9:53 am, "[email protected]" <[email protected] > wrote:

> > Now Rybka is only 8-10 times stronger than GetClub. You give 10 times
> > more time to GetClub and it will play the move that Rybka will play.
>
> How can you make this assertion? Do you have any kind of evidence to
> show that GitClub will *ever* find the same moves as Rybka?

Evidence?!!

Sanny is not a scientist; he is an artist.


-- help bot




  
Date: 10 Jul 2008 17:28:57
From: help bot
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 10, 8:06 am, Sanny <[email protected] > wrote:

> > According to the description by Mr. Guest,
> > it also did not do Q-search--- which is the
> > very same problem we see with Sanny's
> > program, which is causing the program to
> > be even weaker at tactics than humans
> > are.
>
> What is Q search?


Precisely.


--------------------------------------------------------------

We seemed to have a small problem with
the precise spelling of the term, so it was
shortened to just "Q". It's a chess-
programming technique by which tactical
exchanges are "resolved", handing the
result backward to the earlier nodes.

Where there are no tactics, the score of
a given position is just that; but where
there are checks and/or captures, they
are first "solved" and the result is then
passed back. In this way, you would
never score a position as "I'm ahead 0.1
because of my opponent's doubled
pawns" when you are really hanging a
Queen next move, to get out of check.

Chess is 99% tactics, sort of.


-- help bot


  
Date: 10 Jul 2008 08:04:18
From: Sanny
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 10, 6:53=A0pm, "[email protected]" <[email protected] > wrote:
> > Now Rybka is only 8-10 times stronger thanGetClub. You give 10 times
> > more time toGetCluband it will play the move that Rybka will play.
>
> How can you make this assertion? =A0Do you have any kind of evidence to
> show that GitClub will *ever* find the same moves as Rybka?

I regularly play against Rybka and at the moment I find Rybka at 5
seconds plays as good as Easy Level at 50 seconds. So I feel GetClub
is 10 times weaker than Rybka.

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html


   
Date: 10 Jul 2008 11:17:56
From: Guest
Subject: Re: When only 1 legal move, GetClub will not think more.
"Sanny" <[email protected] > wrote in message
news:[email protected]...
>
>I regularly play against Rybka and at the moment I find Rybka at 5
>seconds plays as good as Easy Level at 50 seconds. So I feel GetClub
>is 10 times weaker than Rybka.

It's been mentioned before by many people and you ignored it before, but
I'll try again...

Your reasoning is flawed.

Rybka at 5 seconds per move is not going to play like it does at full time
settings.

For 5 seconds, it barely has time to do anything other than a quick tactical
search. It doesn't have time to bring out its full strength. It's like
comparing a bicycle to a motorcycle by seeing how fast you can back each out
of the parking spot. It's not realistic.


Also, Rybka will likely make much better use of its time than your program
will. Meaning that for Rybka, going from 5 seconds a move to 50 seconds a
move (10 times) will result in a much stronger game than if you gave your
program 500 seconds a move. Going all the way to 3 minutes a move for Rybka
will make it even stronger than if you increased your program's time by the
comparable amount.

The reason is that Rybka is likely to be much more efficient than yours. It
just makes better use of the time. In the factor of 36 difference in time
between a 5 second search and a normal 3 minute search, Rybka would probably
increase its search depth by up to nearly 5 plies.

(I have never played Rybka or looked at its source and I do not know its
search details, but top programs tend to have a search depth growth rate of
around 2.x. Meaning a factor of 2.x in time or speed means one extra ply of
search. This is in significant contrast to the antique programs that had a
growth of around 6.)

And since you are comparing your program to rybka when it's playing a such a
crippled time, your ratings can not be trusted.


If you were to do comparisons with Rybka at the normal 3 minutes a move,
then your results would be a bit more valid. And you could even use the
Technology curve to try and come up with some believable ratings. Provided
you *always* qualified the ratings with how you got them.

But doing Rybka at 5 seconds per move is ridiculous. Getting ratings from
such drastic time differences aren't realistic. And you need lots of games
against multiple oponents to come up with even vaguely believable numbers.



Also, by "10 times weaker", what do you mean... Wouldn't that suggest that
if Rybka has a rating of 3000, then your program would be 300??? (nasty
grin) (Alright, I know the math doesn't actually work quite like that, but
I couldn't resist.)





----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
---= - Total Privacy via Encryption =---


  
Date: 10 Jul 2008 08:02:39
From: Sanny
Subject: Re: When only 1 legal move, GetClub will not think more.
> Indeed. =A0My contention was that MyChess plays "decent enough"
> especially considering the small footprint and age. =A0It is certainly
> not of championship caliber and will likely lose to better human
> players. =A0I do want to match it against GitClub some time just for the
> fun of seeing the resulting game.

Waiting for your games. And lets see how that scores against getclub.

If you know its rating, Then play with Beginner & Easy level and we
will know exact rating of GetClub Chess.

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html


  
Date: 10 Jul 2008 06:53:44
From: [email protected]
Subject: Re: When only 1 legal move, GetClub will not think more.

> Now Rybka is only 8-10 times stronger than GetClub. You give 10 times
> more time to GetClub and it will play the move that Rybka will play.

How can you make this assertion? Do you have any kind of evidence to
show that GitClub will *ever* find the same moves as Rybka?


  
Date: 10 Jul 2008 06:52:40
From: [email protected]
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 10, 12:22=A0am, help bot <[email protected] > wrote:
> On Jul 9, 5:57 pm, "[email protected]" <[email protected]> wrote:
>
> > > MyChess did have two problems though.,,, (etc.)
>
> > Very interesting info about this classic chess program. =A0However,
> > consider this: my compressed DOS binary is about 20k in size, and that
> > includes CGA graphics! =A0The program does a lot, considering its age
> > and its incredibly tiny footprint. =A0I have much admiration for what
> > the programmers of that era were able to do with relatively limited
> > computing resources. =A0In the end MyChess plays a decent enough game
> > for the most part, and even plays by the rules! =A0Sanny might learn
> > from this.
>
> =A0 Apparently, it did not originally play by the
> rules, since you could not promote a pawn
> to a piece unless one of that type had
> already been captured.

This is apparently fixed in the version I have of MyChess so I never
saw this problem. Compare with GitClub where I still am not sure it
plays by the rules.

> =A0 According to the description by Mr. Guest,
> it also did not do Q-search--- which is the
> very same problem we see with Sanny's
> program, which is causing the program to
> be even weaker at tactics than humans
> are.

Indeed. My contention was that MyChess plays "decent enough"
especially considering the small footprint and age. It is certainly
not of championship caliber and will likely lose to better human
players. I do want to match it against GitClub some time just for the
fun of seeing the resulting game.

If I understand correctly --- and I'm sure you can correct any
misconceptions --- lack of quiescence search will cause problems due
to a horizon effect. So that would make the program weaker at tactics
than humans in some situations, but certainly not all, and perhaps not
even most situations.


   
Date: 10 Jul 2008 09:56:02
From: Guest
Subject: Re: When only 1 legal move, GetClub will not think more.
<[email protected] > wrote in message
news:[email protected]...
>
>If I understand correctly --- and I'm sure you can correct any
>misconceptions --- lack of quiescence search will cause problems due
>to a horizon effect. So that would make the program weaker at tactics
>than humans in some situations, but certainly not all, and perhaps not
>even most situations.

A static exchange evaluator will work out the regular attacks & exchanges,
but it will missevaluate the more subtle aspects.

Like pins, xray attacks, the exchanges weakening your position, and so on.

Second order stuff beyond just the material exchanges themselves. Say you
do an exchange involving your king pawn shield in a castled position. A
static exchange wouldn't notice the weakening of your king's security. It
would just evaluate the material balance change.

That does add some error to the evaluation, but often a static exchange is
'good enough' because there will be other reasons for the search to reject
that misevaluated position.

The problems are

1) a static exchange evaluator can actually take as much time as a good
q-search.

2) a static exchange does missevaluate the side effects of the exchanges.
Things that a q-search will usually notice.


In the early days of micro programs, people weren't quite as good at control
the qsearch explosions, so a more predictable static exchange evaluator
seemed a better choice.

(Even in mainframe programs of the early 80s, there were sometimes qsearch
explosions that could cause a 3 minute move to take more than an hour.
Qsearches require some care to avoid that, and you need to enforce your time
controls just in case.)






----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
---= - Total Privacy via Encryption =---


  
Date: 10 Jul 2008 05:06:44
From: Sanny
Subject: Re: When only 1 legal move, GetClub will not think more.

> =A0 According to the description by Mr. Guest,
> it also did not do Q-search--- which is the
> very same problem we see with Sanny's
> program, which is causing the program to
> be even weaker at tactics than humans
> are.

What is Q search?

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html




   
Date: 10 Jul 2008 09:57:00
From: Guest
Subject: Re: When only 1 legal move, GetClub will not think more.

"Sanny" <[email protected] > wrote in message
news:8fe1f7d2-b45d-4eb7-a090-8326b44f38dd@y21g2000hsf.googlegroups.com...

>What is Q search?

It's often called a 'capture search'. Many people do call it a Quiescence
search, but I often don't because I can rarely remember how to spell
quiescence..... (Hence, Q-Search.)

It's what you do when you reach a 'terminate search' node where you would
evaluate it. But since the position isn't 'quiet' or 'calm' (meaning there
are still attacks in progress) you need to extend the search a bit.

Rather than just keep doing full plies with all the moves, you are more
selective about what moves you include.

Usually this is just captures, which is why it's so often called a "capture
search."


The full subject is significantly more complicated than what I just said.
Things like search explosions, other choices besides captures, limiting the
captures included, the move ordering in the Qsearch, and so on can all
significantly improve the quality and speed of the Qsearch.








----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
---= - Total Privacy via Encryption =---


  
Date: 10 Jul 2008 05:04:03
From: Sanny
Subject: Re: When only 1 legal move, GetClub will not think more.
> =A0 Judging from the program's current
> behavior, it "sees" a forced mate coming
> several plies earlier than it used to (if it
> even saw anything at all).
>

Today again the game strength was increased.

Now Easy Level will give you a tough game. I saw your game against
Normal Level that you win by 2 Rooks & Queen threathening on the King.

http://www.getclub.com/playgame.php?id=3DDM22288&game=3DChess

That was a quick Mate. on 31 moves.

Now you can have much strong game even with Easy Level. As now Easy
Level will play stronger than ever.

Now, I feel you will not hear any improvement in GetClubs game as now
all ideas of improvements have exhausted.

Only if you can suggest some way of improvement further that can be
done.

Now Rybka is only 8-10 times stronger than GetClub. You give 10 times
more time to GetClub and it will play the move that Rybka will play.

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html




  
Date: 09 Jul 2008 23:33:06
From: help bot
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 9, 2:12 pm, "Guest" <[email protected] > wrote:

> As a compromise, you could use a set of standardized positions to test
> against. At the VERY LEAST, the 24 Bratko-Kopec test positions. The "Win
> At Chess" test positions are also very popular. The "Albert Silver"
> positions are also popular.
>
> (Note that the ratings estimate derived from the Bratko-Kopec positions are
> known to be bogus. So don't even bother trying to estimate your strength
> from them.)
>
> That way you can report how many correct solutions are found at a setting of
> XYZ on a ABC system. Something reasonably identifying and repeatable.

This could be useful in comparing the
different levels at GetClub to one another,
and to comparing performance in tactics
over time.


> Also, I'm a little suspicious that the program speed is doubling every
> month. That means in 5 months it's 32 times faster than before. This is
> *not*, repeat *NOT*, the same as comparing the play to Rybka or anything
> like that, but a real, genuine faster program (like if you moved to a
> computer that was 32 times faster than it was 5 months ago.)


Note that Sanny's program started out
well below the 1000 USCF level, based on
casual observation. So if he were to add
a whopping 600 points due to repeated
speed-ups, this is not so far-fetched as
you might imagine.

Judging from the program's current
behavior, it "sees" a forced mate coming
several plies earlier than it used to (if it
even saw anything at all).


-- help bot



   
Date: 10 Jul 2008 09:54:42
From: Guest
Subject: Re: When only 1 legal move, GetClub will not think more.

"help bot" <[email protected] > wrote in message
news:[email protected]...
> On Jul 9, 2:12 pm, "Guest" <[email protected]> wrote:
>
>> As a compromise, you could use a set of standardized positions to test
>> against. At the VERY LEAST, the 24 Bratko-Kopec test positions. The
>> "Win
>> At Chess" test positions are also very popular. The "Albert Silver"
>> positions are also popular.
>>
>> (Note that the ratings estimate derived from the Bratko-Kopec positions
>> are
>> known to be bogus. So don't even bother trying to estimate your strength
>> from them.)
>>
>> That way you can report how many correct solutions are found at a setting
>> of
>> XYZ on a ABC system. Something reasonably identifying and repeatable.
>
> This could be useful in comparing the
> different levels at GetClub to one another,
> and to comparing performance in tactics
> over time.

That is why I suggested it.

I thought that if he wasn't willing to do full open testing, then he could
at least take some of the commonly accepted standardized testing positions
and report those results.


I was hesitant to even mention the Bratko-Kopec test's attempt at producing
an elo rating for a program. Trying to get a rating from just 24 positions
isn't really feasible and it's well known the numbers it produces are wrong.
But I figured that if he did look into the BK test, he'd see it anyway.



>
>
>> Also, I'm a little suspicious that the program speed is doubling every
>> month. That means in 5 months it's 32 times faster than before. This is
>> *not*, repeat *NOT*, the same as comparing the play to Rybka or anything
>> like that, but a real, genuine faster program (like if you moved to a
>> computer that was 32 times faster than it was 5 months ago.)
>
>
> Note that Sanny's program started out
> well below the 1000 USCF level, based on
> casual observation. So if he were to add
> a whopping 600 points due to repeated
> speed-ups, this is not so far-fetched as
> you might imagine.

I'm not so sure he's actually talking about doubling of his program's speed.

He seems to be talking more about his program finding the claimed 'right'
rybka answer that is increasing.

That's not quite the same thing. Which is why I made sure I stressed that.

Those kind of improvements are the 'fuzzy' kind that require formal testing
methods to estimate ratings. You can't estimate based on the Tech curve.


For his program to start below 1000 rating is pretty significant... How the
EXPLETIVE can you get a proram to play that poorly on today's hardware??!
Even a program that looks at only material and mobility should be stronger
than that.


> Judging from the program's current
> behavior, it "sees" a forced mate coming
> several plies earlier than it used to (if it
> even saw anything at all).
>
>
> -- help bot
>





----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
---= - Total Privacy via Encryption =---


  
Date: 09 Jul 2008 23:22:49
From: help bot
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 9, 5:57 pm, "[email protected]" <[email protected] > wrote:

> > MyChess did have two problems though.,,, (etc.)
>
> Very interesting info about this classic chess program. However,
> consider this: my compressed DOS binary is about 20k in size, and that
> includes CGA graphics! The program does a lot, considering its age
> and its incredibly tiny footprint. I have much admiration for what
> the programmers of that era were able to do with relatively limited
> computing resources. In the end MyChess plays a decent enough game
> for the most part, and even plays by the rules! Sanny might learn
> from this.

Apparently, it did not originally play by the
rules, since you could not promote a pawn
to a piece unless one of that type had
already been captured.

According to the description by Mr. Guest,
it also did not do Q-search--- which is the
very same problem we see with Sanny's
program, which is causing the program to
be even weaker at tactics than humans
are.


-- help bot




   
Date: 10 Jul 2008 09:53:47
From: Guest
Subject: Re: When only 1 legal move, GetClub will not think more.
"help bot" <[email protected] > wrote in message
news:ae379fae-a953-4aef-b911-6572cf2e44a6@l42g2000hsc.googlegroups.com...
> On Jul 9, 5:57 pm, "[email protected]" <[email protected]> wrote:
>
> Apparently, it did not originally play by the
> rules, since you could not promote a pawn
> to a piece unless one of that type had
> already been captured.

The first version was that way. Maybe the first cp/m shareware version too,
I'm not sure.

That was definetly fixed by later versions. (At what point I don't know.
Just that it was definetly fixed.)

It was just simply a quirk of how he programmed it.

He knew that at the time but did it anyway. He had never written a chess
program before, and probably had read little, so I don't consider it a big
issue for a first version.

Of course, if you are talking about FIDE rules, then it's safe to say few of
the early micro programs played properly. That tended to be added after a
few revisions.

They often didn't do 50 move rule (even though it was simple) or draw by
repetition because that required extra storage.

They also rarely did time control properly.

Of course, later versions of MyChess fixed that, just as the Spracklens did
for Sargon.


> According to the description by Mr. Guest,
> it also did not do Q-search--- which is the

MyChess used a static exchange evaluator. That's a far step above doing
nothing. And there are various levels of sophistication for static exchange
evaluators. I don't know how good MyChess' was.

A static exchange evaluator will take care of the obvious exchanges, but
they do have trouble with pins and xray attacks and so on. The second order
effects.

It's just not as good as a Q-search, but it's better than nothing.


> very same problem we see with Sanny's
> program, which is causing the program to
> be even weaker at tactics than humans
> are.

That would depend on where the tactical errors are being made.

People have mentioned shallow tactical errors and that would suggest it's
neither a q-search or horizon problem.

Still, I guess if he cut off the search in the middle of an exchange, that
would certainly cause problems.








----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
---= - Total Privacy via Encryption =---


  
Date: 09 Jul 2008 14:57:44
From: [email protected]
Subject: Re: When only 1 legal move, GetClub will not think more.

>
> MyChess did have two problems though.,,, (etc.)

Very interesting info about this classic chess program. However,
consider this: my compressed DOS binary is about 20k in size, and that
includes CGA graphics! The program does a lot, considering its age
and its incredibly tiny footprint. I have much admiration for what
the programmers of that era were able to do with relatively limited
computing resources. In the end MyChess plays a decent enough game
for the most part, and even plays by the rules! Sanny might learn
from this.


  
Date: 09 Jul 2008 09:33:23
From: Sanny
Subject: Re: When only 1 legal move, GetClub will not think more.
> It's the same situation as with testing... you need lots of games to be a=
ble
> to get a reasonable confidence in the actual strengths of the programs.
>
> Yes I know that is the way tournaments are done. =A0I'm not disagreeing. =
=A0I'm
> saying that you can't get good statistical results with that few games an=
d
> whatever winner you do get is more 'show' than fact.

Thats true it is very difficult to find out correct strength of
program. And GetClub is doubling in speed every month it is very
difficult to estimate its strengths.

Now, Even Easy Level gives good competition. Have you ever played at
GetClub? Play a game with easy level and tell me how you find it.

To swim you have to dive in river.
To play you have to start a Chess competition.

Play Chess at: http://www.GetClub.com/Chess.html

So Just start a game with Easy Level and see how good GetClub plays.

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html



   
Date: 09 Jul 2008 13:12:18
From: Guest
Subject: Re: When only 1 legal move, GetClub will not think more.
>"Sanny" <[email protected]> wrote in message
>news:cbf99adb-9342-42a4-a886->[email protected]...
>> It's the same situation as with testing... you need lots of games to be
>> able
>> to get a reasonable confidence in the actual strengths of the programs.
>>
>> Yes I know that is the way tournaments are done. I'm not disagreeing. I'm
>> saying that you can't get good statistical results with that few games
>> and
>> whatever winner you do get is more 'show' than fact.
>
>Thats true it is very difficult to find out correct strength of
>program. And GetClub is doubling in speed every month it is very
>difficult to estimate its strengths.

Yes, it is difficult. That's why you need automated testing etc. And
playing on the free intenet chess servers where you can get reliable test
results.

Until you do that, you and your program are a joke. Nobody will take it any
more seriously than if you said you had a dog that could talk and sing
saprano.

You can get away without serious testing if you do chess programming as a
hobby, with nobody else caring. But what you are doing (trying to be
commerical and spouting ratings) requires serious testing.

As a compromise, you could use a set of standardized positions to test
against. At the VERY LEAST, the 24 Bratko-Kopec test positions. The "Win
At Chess" test positions are also very popular. The "Albert Silver"
positions are also popular.

(Note that the ratings estimate derived from the Bratko-Kopec positions are
known to be bogus. So don't even bother trying to estimate your strength
from them.)

That way you can report how many correct solutions are found at a setting of
XYZ on a ABC system. Something reasonably identifying and repeatable.

But test positions are no substitute for full testing. Just a convenint way
to do some quickie tests.


Also, I'm a little suspicious that the program speed is doubling every
month. That means in 5 months it's 32 times faster than before. This is
*not*, repeat *NOT*, the same as comparing the play to Rybka or anything
like that, but a real, genuine faster program (like if you moved to a
computer that was 32 times faster than it was 5 months ago.)


Getting back to the doubling of performance every month...

There is a thing called the "Technology curve" that's about how much ratings
increase you get for a doubling in computer power. (Much the research talks
about the strenght as compared to depth of search. But the data can be
converted.)

The is quite a bit of debate as to the actual shape of the curve. Most
though agree that it's not a straight line and is indeed a curve. That its
growth slows drastically on the faster systems (or deeper searches.)

Part of the problem is that most of the results reported are based on
self-play to different depths, rather than full scale games against a
variety of opponents.

In the old days, on slow hardware and when searches were usually 4-7 ply in
the middle game, people usually said an extra ply of search was worth about
200 ratings points.

However, most agree that is no longer true. The deeper you search (or the
better your program is to begin with), the less effect the increased power
will have on your ratings.


For example, based on Szabo & Szabo's testing, a factor of 32 power can give
you about 600 rating points if your program is really really bad to begin
with.

If your program already has a rating of 1500, then the same increase in
power will give you about 400 points.

If your program is already at 2000, then the same factor of 32 performance
increase will give you maybe 250 points increase.

And that's with doubling the computer speed 5 times (2^5=32). And again,
this is *not* the kind of 'doubling of performance' that you would get by
comparing your program's play to Rybka. This is REAL program speed
doublings.

But those are just rough estimates based on self play.

Real results will vary quite a bit. The only way to know what the ratings
change really is is to do serious testing with lots of games and a wide
range of opponents.


Again, if this was a hobby, you could get away with casual testing. Lots of
people do that.

But if you are trying to be serious, and be commerial, and spout ratings
levels, then you need to be serious about your testing.






>Now, Even Easy Level gives good competition. Have you ever played at
>GetClub? Play a game with easy level and tell me how you find it.
>
>To swim you have to dive in river.
>To play you have to start a Chess competition.
>
>Play Chess at: http://www.GetClub.com/Chess.html
>
>So Just start a game with Easy Level and see how good GetClub plays.
>
>Bye
>Sanny
>
>Play Chess at: http://www.GetClub.com/Chess.html
>




----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
---= - Total Privacy via Encryption =---


   
Date: 09 Jul 2008 13:25:24
From: Kerry Liles
Subject: Re: When only 1 legal move, GetClub will not think more.
Doubling in strength every month... lets see: after 6 months that would be 0
* 0 * 0 * 0 * 0 * 0 * 0 = zero


"Sanny" <[email protected] > wrote in message
news:cbf99adb-9342-42a4-a886-7c93da213679@k13g2000hse.googlegroups.com...
Thats true it is very difficult to find out correct strength of
program. And GetClub is doubling in speed every month it is very
difficult to estimate its strengths.





  
Date: 09 Jul 2008 06:46:46
From: [email protected]
Subject: Re: When only 1 legal move, GetClub will not think more.
> MyChess was written by Dave Kittenger. =A0In 78 he was living in Alaska a=
nd
> was one of the top players in the entire state. =A0

MyChess was really not bad for its time (and you can still get a
working copy for DOS emulation). It plays at least a reasonable game
although it gets pretty slow when set to "higher" play levels. I
really ought to match it against GitClub; that would be most
interesting.


   
Date: 09 Jul 2008 10:59:40
From: Guest
Subject: Re: When only 1 legal move, GetClub will not think more.
><[email protected]> wrote in message
>news:bafb31b9-caeb-4fa5-9e77->[email protected]...
>> MyChess was written by Dave Kittenger. In 78 he was living in Alaska and
>> was one of the top players in the entire state.
>
>MyChess was really not bad for its time (and you can still get a
>working copy for DOS emulation). It plays at least a reasonable game
>although it gets pretty slow when set to "higher" play levels. I
>really ought to match it against GitClub; that would be most
>interesting.

MyChess was a pretty good first attempt. Far better than Spracklen's Sargon
was. Too bad he never released the source. I tried contacting him once
about it, but all I could find was a postal address and he didn't write
back.

MyChess was also faster than Sargon. About 10 times faster.

It had a better evaluator and move ordering too.

MyChess did have two problems though.

First, it used a static exchange evaluator instead of a Qsearch. Although
they were popular in the early micro days, they do have a number of tactical
problems and can cause serious misevaluations.

The second problem was that the way MyChess stored the board, you couldn't
promote a Queen to a piece that hadn't been captured. Meaning if you still
had both rooks, you couldn't promote to a rook.

MyChess used a very odd board layout, where each of the 32 pieces had a
fixed location in an array and the array held the piece square number. This
is in contrast to other programs that have an array where each square hold
the type of piece on that square.

The promotion issue was fixed in later versions, of course.






----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
---= - Total Privacy via Encryption =---


  
Date: 08 Jul 2008 17:31:42
From: help bot
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 7, 9:19 pm, "Guest" <[email protected] > wrote:

> > I don't recall having much in the way of choice,
> > until I got a PC, that is. They had a cartridge--
> > you could buy it, or not, your choice. :>D
>
> Oh, there were lots of chess programs in the early 80s. I remember getting
> computer magazines around 80 (at the earliest) and onward, and there were
> quite a few.
>
> I don't remember them all, but here are a few from that time period:
>
> MyChess, MicroChess, Sargon 1 & 2, Sfinks (of course), Micro-Chess (not
> Peter Jennings), Video Chess, 8080 Chess. Chess 0.5 (Pascal source, not
> much of a chess program.)

I didn't have a PC quite that early. Maybe some
of these were for the Radio Shack TRS/80, which
I never had?


> And those are just off the top of my memory from the period 78-81ish. And
> there were dozens more competing in tournaments. The odds are good at least
> a few of them were sold commercially too (like Sargon & MyChess & Sfinks
> were.)

The name Sargon is the frist to strike a bell.
But I'm thinking of Sargon IV... it was so long ago.


> My little unpopular micro (RadioShack Color Computer) had... ChessD,
> MicroChess, Cyrus, VoxChess, Chess, and at least one more that I can't
> remember the name of. And that was an unpopular system.

My guess is it wasn't a good deal; that Radio
Shack was stiffing its customers. Either that,
or maybe it was just that Atari had much
better games at the time.


> Something popular like the C64 or Apple ][ would have at least double that.

When the Commodore64 first came out, it
undercut Atari by a huge margin, and added
16k to boot. However, the selection of
software was terrible, because it was so new.


> For at least the past 10 years, he didn't *have* a copy.

Everything I mentioned happened long
before this.


> I scanned them in the printed copies and tried to OCR them, but without much
> luck. (Have you ever tried to OCR Fortran???

Punched cards... we had punched cards back
then, easily read by a machine. The program
consisted of a stack of punched cards, all in
their proper sequence. You transported the
programs in your satchel, stored in the trunk
of your Ford Falcon or Chevy Bel Aire.


> >> He's right. It's not reasonable to port CrayBlitz to another system.
>
> > When a person is "right", he does not ever
> > need to contradict himself the way BH did.
>
> > For instance, I noticed that in many of the
> > old discussions he wanted/needed to have
> > things both ways. If they were talking about
> > his various chess programs, BH might write
> > something to the effect that one was in
> > essence the same as another, only on
> > different hardware. But if someone else--
> > say, one of those who wanted to know how
> > "smart" the program was vs. how "fast" was
> > a Cray-- would ask an annoying question,
> > Mr. Hyatt would simply contradict himself.
> > This lent the impression that he was not
> > merely wrong, but averting the truth.

> Not quite sure what you are saying there.

Self-contradiction is where a person says
one thing, but then turns around and says
the opposite, when it suits his whim.

For instance, Dr. IMnes -- a famous nearly-
an-IM -- will support the sinister Larry Parr in
"trying and convicting" the president of FIDE
of murdering a journalist based on purely
circumstantial evidence, but then turn around
on a whim and insist that this sort of thing is
un-American, when applied to somebody
else.


> The program itself had some chess knowledge. Quite a bit. (I didn't
> examine the algorithms that closely, so I can't say what.)
>
> Except for the parallel processing stuff, there probably wasn't much that
> was truely unique or unusual. But what was there was very often very Cray
> specific. Things were often chosen and coded to make the Cray happy.
>
> He did depend heavily on the Cray's raw performance to work as well as it
> did.
>
> It was kind of like how Ken Thompson gave up on regular computers and built
> Belle. Hyatt gave up on regular computers and used a Cray.

There is a slight difference there, as few
people I know of could afford a Cray. Let
me see... Sam Walton... Bill Gates...
Warren Buffet... the Prince of Saudi
Arabia... Elvis. Michael Jackson, after
his Thriller album maybe.


> You can't easily seperate the Cray out of CrayBlitz. It's not called "Blitz
> on Cray".
>
> It would take a massive rewrite to get the Cray out of it and make it
> suitable for a regular computer. Not just a translation, but you'd have to
> rewrite everything that depended on the Cray hardware.

Indeed, this contradicts what Mr. Hyatt
himself wrote, and I suspect that you are
right and that he was attempting to carry
forward, so to speak, his fame and glory
from CrayHeaven-to-reality.


> He did have a portable version, but it was basically just the regular
> Fortran version doing things sequentially. It was still set up for vectors
> and so on, it was just done sequentially on a regular computer. And as a
> result, it ran very slowly.
>
> Figure a reduction by 10 for regular hardware. Probably not unreasonable.
> That would cut about two plies off its search.

If the chess-knowledge was really good,
chopping off a bit of Cray-speed would not
have all that much impact, except maybe
in competitions against other top-notch
contenders.


> >> The reality is, the Cray architecture is highly unsuited to computer
> >> chess.
> >> Chess does not vectorize well, at all. That he managed to make it work
> >> as
> >> well as he did is a testiment to his & Nelson's programming skills.

Yet it would seem to indicate poor
judgment, if what you say just above is
true. But all these apologies come off
as rather unconvincing, given the fact
that using the Cray, they managed to
win the world computer championship;
without the Cray... nothing.


> Considering the cost of Cray time, he would have needed to sell his house to
> pay for *one* tournament.

Exactly. So why pretend having free access
was not an advantage? Why deny the speed
and power of the Cray? Why not look at the
facts, like how /with/ the Cray, BH won, but
without the Cray, he never has despite his
long involvement. Game losses were blamed
on "bad pairings" rather than "bad moves", etc.
It all fits a certain pattern, reminiscent of a
famous American world chess champion.


> In the old days, it wasn't easy to do a chess program.

Even today, it's not easy if you (i.e. Sanny)
are unwilling to do some research. Chess
has a lot of special moves, like /en passant/,
castling and pawn promotion. All the pieces
have their own ways of moving, and of course
the King is not allowed to do things that the
other pieces can do. Victor Kortchnoi once
asked the arbiter if he could legally castle;
how is Sanny to know such things, if a
player two thousand rating points higher
does not know them? : >D


-- help bot


   
Date: 08 Jul 2008 22:44:13
From: Guest
Subject: Re: When only 1 legal move, GetClub will not think more.
"help bot" <[email protected] > wrote in message
news:f25a827f-ba8a-46ca-bdf9-56114c3e7c72@m36g2000hse.googlegroups.com...
> On Jul 7, 9:19 pm, "Guest" <[email protected]> wrote:
>
>> MyChess, MicroChess, Sargon 1 & 2, Sfinks (of course), Micro-Chess (not
>> Peter Jennings), Video Chess, 8080 Chess. Chess 0.5 (Pascal source, not
>> much of a chess program.)
>
> I didn't have a PC quite that early. Maybe some
> of these were for the Radio Shack TRS/80, which
> I never had?

MyChess was for z80 systems, and later x86 systems (cp/m & dos).

MicroChess was for lots of processors. There were several versions of
Microchess, with the later ones being much better.

Sfinks would have been z80 in those days, with the PC version coming out in
83?

(For the Z80 stuff, they were probably CP/M, so were fairly portable among
computer brands.)

Chess 0.5 was in Pascal and was published in 1978. It was a lousy chess
program, but it served as a basis for many others. Just like Sargon 1 did.

Sargon 1 was z80 and almost any OS. It was also ported to the 6502 and 6809
processors. Probably the x86 too but I can't remember for sure.

8080 chess was just for the Sol-20 with an 8080 processor. All the way back
to 1977.

But those were just off the top of my head. I'm sure there were others that
I can't remember in the -1981 days.

After 1981 or so, there were lots of chess programs.


>> And those are just off the top of my memory from the period 78-81ish.
>> And
>> there were dozens more competing in tournaments. The odds are good at
>> least
>> a few of them were sold commercially too (like Sargon & MyChess & Sfinks
>> were.)
>
> The name Sargon is the frist to strike a bell.
> But I'm thinking of Sargon IV... it was so long ago.

Sargon became famous when it was written in 3 months (before the Spracklens
even owned a computer) and won the 1978 West Coast Computer Faire chess
tournament. (The guy who organized that later wrote the Atari 2600 chess
game.)

MyChess was written by Dave Kittenger. In 78 he was living in Alaska and
was one of the top players in the entire state. So he wrote his own chess
program. It supposedly had a very 'human' style play. These programs later
evolved and were put into the Novag chess computers, such as the Novag
Savant. Later descendants were the Connie & Super Connie.



>
>
>> My little unpopular micro (RadioShack Color Computer) had... ChessD,
>> MicroChess, Cyrus, VoxChess, Chess, and at least one more that I can't
>> remember the name of. And that was an unpopular system.
>
> My guess is it wasn't a good deal; that Radio
> Shack was stiffing its customers. Either that,
> or maybe it was just that Atari had much
> better games at the time.

In those days, it was always a gamble as to which computer to buy. It
depended on what you wanted etc.

The RadioShack stuff usually wasn't that bad of stuff, but since you had to
buy at a Radio Shack store, they weren't as accessible as many of the other
brands, such as the Apple ][ and the Commodore 64 that you could buy at lots
of stores and even mail order.

Even the TI 99/4 and Atari computers were much more widely available.


I didn't really have much to complain about the CoCo. It's processor was
significantly more powerful than the 1mhz & 2mhz 6502 that other systems
used.

I don't regret not getting an Apple and I certainly didn't want a C64.



>> Something popular like the C64 or Apple ][ would have at least double
>> that.
>
> When the Commodore64 first came out, it
> undercut Atari by a huge margin, and added
> 16k to boot. However, the selection of
> software was terrible, because it was so new.

That was always true of a new computer.

There was very little concern about 'backward portability' in those days.

With the C64, you did have some compatability, though. They were partially
compatable with the Vic-20, and could run some of the PET programs in
'compatability' mode.




>> For at least the past 10 years, he didn't *have* a copy.
>
> Everything I mentioned happened long
> before this.

And that's why I pointed out that in the days of when he was still doing
CrayBlitz, he certainly wouldn't have released the source.

And in those few years after CrayBlitz while the source was still available,
he certainly wasn't in the mood to release it.

After that, there was no way he could release it.



>> I scanned them in the printed copies and tried to OCR them, but without
>> much
>> luck. (Have you ever tried to OCR Fortran???
>
> Punched cards... we had punched cards back
> then, easily read by a machine. The program

Not easily read if you were working with a *printed* copy from the
printer.... Which is what I had to deal with.

The tournament organizers required a printed copy, so that's what you had to
provide.


> consisted of a stack of punched cards, all in
> their proper sequence. You transported the
> programs in your satchel, stored in the trunk
> of your Ford Falcon or Chevy Bel Aire.

I know what punched cards are.

I never used them, but I know what they are.

Cray Blitz was never on punch cards. By that time, they had definetly
fadded into the past.

The original Blitz was probably first done on punch cards, though.


>> >> He's right. It's not reasonable to port CrayBlitz to another system.
>>
>> > When a person is "right", he does not ever
>> > need to contradict himself the way BH did.
>>
>> > For instance, I noticed that in many of the
>> > old discussions he wanted/needed to have
>> > things both ways. If they were talking about
>> > his various chess programs, BH might write
>> > something to the effect that one was in
>> > essence the same as another, only on
>> > different hardware. But if someone else--
>> > say, one of those who wanted to know how
>> > "smart" the program was vs. how "fast" was
>> > a Cray-- would ask an annoying question,
>> > Mr. Hyatt would simply contradict himself.
>> > This lent the impression that he was not
>> > merely wrong, but averting the truth.
>
>> Not quite sure what you are saying there.
>
> Self-contradiction is where a person says
> one thing, but then turns around and says
> the opposite, when it suits his whim.

(shaking head) I know what self contradicting is. You don't need to get
that tone.

I was saying that I didn't know what the point was... What subject was he
contradicting himself.



>> The program itself had some chess knowledge. Quite a bit. (I didn't
>> examine the algorithms that closely, so I can't say what.)
>>
>> Except for the parallel processing stuff, there probably wasn't much that
>> was truely unique or unusual. But what was there was very often very
>> Cray
>> specific. Things were often chosen and coded to make the Cray happy.
>>
>> He did depend heavily on the Cray's raw performance to work as well as it
>> did.
>>
>> It was kind of like how Ken Thompson gave up on regular computers and
>> built
>> Belle. Hyatt gave up on regular computers and used a Cray.
>
> There is a slight difference there, as few
> people I know of could afford a Cray. Let
> me see... Sam Walton... Bill Gates...
> Warren Buffet... the Prince of Saudi
> Arabia... Elvis. Michael Jackson, after
> his Thriller album maybe.

He couldn't afford a Cray either.

Cray *donated* the computer time for free. It was good publicity for them.

Just like most universities donated their mainframe time.

Just like Control Data Corp did for Slate & Atkin for their 'Chess' line of
programs. It was good publicity.

And so on.

You don't have to buy a Cray (or other mainframe or super computer) to use
it. And if you can do some fast talking, you don't even need to rent it.
You can get the computer time for free.

Things are a little differen these days, but not a whole lot.

Hyatt still gets donated computer time from AMD & Intel and his university.

Hsu got IBM to actually build a super computer to play chess.

And so on.


>> You can't easily seperate the Cray out of CrayBlitz. It's not called
>> "Blitz
>> on Cray".
>>
>> It would take a massive rewrite to get the Cray out of it and make it
>> suitable for a regular computer. Not just a translation, but you'd have
>> to
>> rewrite everything that depended on the Cray hardware.
>
> Indeed, this contradicts what Mr. Hyatt
> himself wrote, and I suspect that you are
> right and that he was attempting to carry
> forward, so to speak, his fame and glory
> from CrayHeaven-to-reality.

Without knowing exactly what he said, I can't really comment on what he did
or did not say.

But I wouldn't be too surprised if you are misreading or are taking
something out of context.



>> He did have a portable version, but it was basically just the regular
>> Fortran version doing things sequentially. It was still set up for
>> vectors
>> and so on, it was just done sequentially on a regular computer. And as a
>> result, it ran very slowly.
>>
>> Figure a reduction by 10 for regular hardware. Probably not
>> unreasonable.
>> That would cut about two plies off its search.
>
> If the chess-knowledge was really good,
> chopping off a bit of Cray-speed would not
> have all that much impact, except maybe
> in competitions against other top-notch
> contenders.

Cutting off up to two plies is not trivial. And your suggestion that it
"would not have all that much impact" clearly shows you aren't familiar with
computer chess programming.

And that is just going from the Cray to a regular system.

And that doesn't take into account that the Crays were parallel computers.
In the later days, he was using lots of cores. I don't know how many (I'm
not an expert on Cray's lines and I don't know what Hyatt actually used,
beyond whatever Cray offered), but at least 8, and probably more than that
on the later versions of the Crays.

That loss of parallel processing is going to hurt plenty. Going from 8
processors to just one is going to cost you at least another full ply.

We are now talking about the loss of perhaps 3 plies when a middle game
search might be 10 or 11 plies. (Just a guess because it would depend on
what time period we are talking about. In 1985 he was doing 8 plies or so
in middle game. Later versions were faster, so 10 ply or so could be
reasonable.)

That's not a trivial loss in search depth.

It's going to have a heck of an effect on the quality of play.

And in computer vs. computer play, the program that searches deepest (& has
the best q-search) is almost always going to win.


In my previous post I didn't even mention the parallel search stuff because
I was concentrating on just the Cray - > normal cpu aspects.



>> >> The reality is, the Cray architecture is highly unsuited to computer
>> >> chess.
>> >> Chess does not vectorize well, at all. That he managed to make it
>> >> work
>> >> as
>> >> well as he did is a testiment to his & Nelson's programming skills.
>
> Yet it would seem to indicate poor
> judgment, if what you say just above is

Poor judgement in what way?

He had the worlds fastest computers to play with.

Even in scalar mode, the Cray was pretty darn fast. Just not as fast as it
was in vector mode.

And he & Nelson *were* able to come up with new data structures and
algorithms to take advantage of the unusual hardware.

The only reason he stopped using the Crays was because Cray was falling on
hard times and losing interest in chess and Hyatt had trouble getting enough
computer time even for testing. They were often bumping him off the faster
systems to the slower systems during tournaments.



> true. But all these apologies come off
> as rather unconvincing, given the fact
> that using the Cray, they managed to
> win the world computer championship;
> without the Cray... nothing.

Hyatt lost the world championship when the purpose built hardware came
around. (HiTech, ChipTest, DeepThought & DeepBlue.)

That was about the same time Cray was having financial issues and was
loosing interest in chess.

That was about when he decided to start on an entirely new program that
wouldn't depend on any special hardware. Crafty.



>
>> Considering the cost of Cray time, he would have needed to sell his house
>> to
>> pay for *one* tournament.
>
> Exactly. So why pretend having free access
> was not an advantage? Why deny the speed
> and power of the Cray? Why not look at the

I seriously doubt Hyatt ever dismissed the benefit of the Cray.

He probably said something like: "The cray is very fast and it helps.
Without it I would have to totally rewrite the program. It would be strong,
very strong, but would not be as fast as CrayBlitz, but I believe it would
still be as strong."

That's not meant to be a quote or anything. Just a guess as to what he
probably said.


And as for it being free... as I said above, a lot of people managed to get
free computer time to play chess. CDC & Cray both donated super computer
time. A few times, Cray was running both Slate's NuChess program and
Hyatt's CrayBlitz in the same facility on similar hardware. They were
donating time for two competing programs.


> facts, like how /with/ the Cray, BH won, but
> without the Cray, he never has despite his
> long involvement. Game losses were blamed

As I have said, he's not the best chess programmer.

And at his age, you should be surprised he's still around and productive.

He's been in computer chess 40 years. (I originally said 40 but thought I
was wrong and corrected myself. It turns out I was right because he said he
did start in 68. But from what I can see, he didn't really become very
active and productive until the mid 70s. So 30 years of active computer
chess and 40 years of computer chess.)


> on "bad pairings" rather than "bad moves", etc.
> It all fits a certain pattern, reminiscent of a
> famous American world chess champion.

Many of the chess tournaments were poorly organized. And I'm not talking
about Hyatt's losses or anything like that.

You'd have a dozen programs (some of which are brand new) all competing in a
tournament where every single point counted.

There were often only 4 or 5 rounds, and that's not really enough to judge
the relative quality of play among programs. The error statistics are too
far apart.

There just weren't enough games played to actually say who had the stronger
program.

And I'm not talking about Hyatt at all. I'm talking about all the programs.
A lot of those tournaments were a lot more show than any realistic test to
determine who was stronger.

It's the same situation as with testing... you need lots of games to be able
to get a reasonable confidence in the actual strengths of the programs.


Yes I know that is the way tournaments are done. I'm not disagreeing. I'm
saying that you can't get good statistical results with that few games and
whatever winner you do get is more 'show' than fact.



>> In the old days, it wasn't easy to do a chess program.
>
> Even today, it's not easy if you (i.e. Sanny)
> are unwilling to do some research. Chess
> has a lot of special moves, like /en passant/,
> castling and pawn promotion. All the pieces
> have their own ways of moving, and of course
> the King is not allowed to do things that the
> other pieces can do. Victor Kortchnoi once
> asked the arbiter if he could legally castle;
> how is Sanny to know such things, if a
> player two thousand rating points higher
> does not know them? :>D
>
>
> -- help bot
>




----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
---= - Total Privacy via Encryption =---


  
Date: 07 Jul 2008 20:48:00
From: help bot
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 6, 1:42 pm, "Guest" <[email protected] > wrote:

> > Rybka is world best So I play against it. I find it interesting that
> > how number of moves taken by Rybka is increasing every week.
>
> But it's not representative.
>
> Your testing method is flawed. Seriously flawed.
>
> First, you are tuning your program to play gainst Rybka. That says very
> little about how it would play against others or people.

This is where the analysis errs-- big-time; the
facts is, Rybka is so strong that "tuning" his
program to improve results against it /would/
(you are mistaken to assume he is capable
of such a feat) say *plenty* about how well it
should do against others.


> Let's pretend that Rybka has a major weakness...

In fact, if you disable or do not install an
openings book, it does: a very strong
preference for Nc3/...Nc6, blocking its
own c-pawn and thus cramping its own
Queen.


> Why do you think people keep telling you to do your testing against a wide
> range of other programs? GNU Chess, MicroMax, Crafty, Fruit, and so on?

Those programs would beat GetClub like
carrots. The actual recommendations tend
to focus on playing other weak programs or
humans of known mediocre strength.


> > 3 months back Rybka used to win Easy Level in 25 moves.
> > 2 months back it started taking 30 moves / average.to win
> > 1 month back it started taking 35 moves / average to win
> > Now it takes 40+ moves to win the Easy Level.
>
> All that says is that you are slowly making your program less terrible. Or
> that it's faster and just isn't making as many tactical blunders.
>
> That doesn't say anything how well it really plays real games.

Wrong. If Sanny's numbers are valid (which
would surprise me), then it shows that an
opponent of fixed strength is having to work
harder to win-- that his program truly is
improving. The problem is that Sanny's
testing methods are more-or-less random.


> And you can *NOT* just do one or two games and assume that your program is
> better now than it was before. Testing takes at least a couple dozen games.
> Preferably HUNDREDS to be able to get a wide range of games. With either
> random books or some sort of even randomness so the programs wont play the
> same sequence of moves over & over.

What I said! (Hey, why is it that when
you finally get something right, it turns
out to have been plagiarized from me?)
; >D


> The interactions between the chess heuristics and the search algorithms can
> cause all sorts of unexpected side effects. Weaker play in some areas but
> stronger play in others.
>
> Lets say you improve your pawn structure, but in doing so, you cause the
> program to like 'bad' bishops. Unless he happens to play a game where the
> 'bad' bishops are a factor, he wouldn't even notice the problem.
>
> Unless you do enough games to get a good spread, you can't reasonably say
> whether a change is good or bad as a whole.
>
> I would say at least 20 games against several opponents, and preferably a
> hundred games or more against at least 4 different programs, including your
> previous program. And make sure you have some randomness somewhere so it
> wont play the same opening lines over & over.
>
> Automate it so the tests can be done without human participation, and just
> let it play for a couple days or a week, and then decide whether the change
> is good or bad.

I expect Sanny does not have a clue how
to automate such testing... and is in dire
need of help here.

I prefer the idea of testing it on FICS, as
that way everybody can see the actual
results for themselves, and not have to
rely on biased sources (ahem) to report
the results.


-- help bot


   
Date: 08 Jul 2008 09:34:58
From: Guest
Subject: Re: When only 1 legal move, GetClub will not think more.
"help bot" <[email protected] > wrote in message
news:4ec4fe49-13eb-426d-836e-6b1b5f55f8db@e53g2000hsa.googlegroups.com...
> On Jul 6, 1:42 pm, "Guest" <[email protected]> wrote:
>
>> > Rybka is world best So I play against it. I find it interesting that
>> > how number of moves taken by Rybka is increasing every week.
>>
>> But it's not representative.
>>
>> Your testing method is flawed. Seriously flawed.
>>
>> First, you are tuning your program to play gainst Rybka. That says very
>> little about how it would play against others or people.
>
> This is where the analysis errs-- big-time; the
> facts is, Rybka is so strong that "tuning" his
> program to improve results against it /would/
> (you are mistaken to assume he is capable
> of such a feat) say *plenty* about how well it
> should do against others.

Sorry, but professional chess programmers disagree. Firmly. These people
have 10 or 20 years programming chess, sometimes doing it to earn a living.
They've learned the hardway about the dangers of insufficient testing or
poor testing methods.

(No, I'm not talking about me. I'm passing on advice that I have been given
by those people. I have written chess programs off & on for 20 years, but
they were never major programs. That was never my goal. I just had fun
programming them.)

You do not *ever* depend on one or even two specific programs as your main
opponents.

It can have the side effect of making it a bit stronger in the process, but
it will also have significant problems. Unbalanced play, tendancies to try
moves that will delay the specific style of play it expects. And so on.

Its the same basic reasoning why you use programs of similar strength. That
way you can win semi-regularly, where as always loosing to a stronger
program would tend to result in your program being trained to delay the loss
instead of going for a win.

As for whether or not he's capable of tuning it... That's simply an aspect
of using one main program (or human opponent) as your primary test bed. If
you make changes and it plays worse, you try something else. That's tuning
it.

That's why you need so many programs (or people) during the tuning phase.





>> Why do you think people keep telling you to do your testing against a
>> wide
>> range of other programs? GNU Chess, MicroMax, Crafty, Fruit, and so on?
>
> Those programs would beat GetClub like
> carrots. The actual recommendations tend
> to focus on playing other weak programs or
> humans of known mediocre strength.

You do need a wide range of *comparable* chess programs, yes.

But I figured he probably wouldn't be too interested in play TSCP or the
ChessProgramming wiki's new free program or other comparable programs.

He'd be more intererested in Fruit & crafty. Even GNU Chess has a
reasonable reputation and is well known. (Although it's not as strong a
player, it is at least well known.)

I figured MicroMax 4.8 (which is under 2k of C source) would be pushing what
he'd be willing to look at. In spite of it being so small, it's
surprisingly strong due to the tactical nature of chess and the speed of
today's systems.


>> > 3 months back Rybka used to win Easy Level in 25 moves.
>> > 2 months back it started taking 30 moves / average.to win
>> > 1 month back it started taking 35 moves / average to win
>> > Now it takes 40+ moves to win the Easy Level.
>>
>> All that says is that you are slowly making your program less terrible.
>> Or
>> that it's faster and just isn't making as many tactical blunders.
>>
>> That doesn't say anything how well it really plays real games.
>
> Wrong. If Sanny's numbers are valid (which
> would surprise me), then it shows that an

By your own admission here and in other messages, you *don't* believe his
numbers are valid.

So why are you disagreeing now?

> opponent of fixed strength is having to work
> harder to win-- that his program truly is
> improving. The problem is that Sanny's
> testing methods are more-or-less random.

I didn't say it wasn't slowly improving. In fact, I did say that it could
be "less terible" and "isn't making as many tactical blunders".

I said that him comparing how many moves it takes to loose against Rybka
said nothing about how well it actually played. That his program could
simply be learning how to delay the loss.

Delaying a loss doesn't mean you have any idea how to win.

It's like the classic horizon effect in computer chess. You are in a
position where you are about to loose your queen, but you can block it by
moving a pawn. When that gets captured, you can block it with another pawn.
And then a rook. And so on. Pretty soon, you've lost the queen and the
rooks and the pawns.

You delayed the loss but are now in a worse position than you were before.



>
>
>> And you can *NOT* just do one or two games and assume that your program
>> is
>> better now than it was before. Testing takes at least a couple dozen
>> games.
>> Preferably HUNDREDS to be able to get a wide range of games. With either
>> random books or some sort of even randomness so the programs wont play
>> the
>> same sequence of moves over & over.
>
> What I said! (Hey, why is it that when
> you finally get something right, it turns
> out to have been plagiarized from me?)
> ;>D

I've said that several other times in here.

You aren't exactly the star in this group, so why would you think I'd repeat
anything you said?

Even joking it's insulting for you to suggest I'd repeat you.


>> Automate it so the tests can be done without human participation, and
>> just
>> let it play for a couple days or a week, and then decide whether the
>> change
>> is good or bad.
>
> I expect Sanny does not have a clue how
> to automate such testing... and is in dire
> need of help here.

Possibly not.

But I suspect it's probably more along the lines of not wanting to.

If he did do automatic testing, which would allow lots of real games and
comparable time controls etc., he would no longer be able to say he believed
his program was now at 4300 elo at the easy level....



> I prefer the idea of testing it on FICS, as
> that way everybody can see the actual
> results for themselves, and not have to
> rely on biased sources (ahem) to report
> the results.

That's something I think he *definetly* does not want to do.






----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
---= - Total Privacy via Encryption =---


  
Date: 07 Jul 2008 15:55:42
From: help bot
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 6, 11:52 am, "Guest" <[email protected] > wrote:

> But I admit I am curious about how it would compare to the older programs
> (sargon 2) or older harrdware.

I still have a couple of older computers lying
around, gathering dust. But I never seem to
get around to putting them back together, and
I fear the nightmare of those old OSs -- which
have left a bad taste in my mouth, even today.

Testing memory...


> You must be talking about one of the first Chess Challengers in the 70's or
> something. Some of those were pretty limited hardware. Sometimes as little
> as 2k of ROM.

The Boris Diplomat, actually.


> And there were a heck of a lot more chess programs for the 8 bit systems, at
> least until the PC really took off in the mid 80s.

I don't recall having much in the way of choice,
until I got a PC, that is. They had a cartridge--
you could buy it, or not, your choice. : >D


> >> The only reasons I've mentioned it are because of the hash table test
> >> Nelson
> >> did (which is very relevant to the micro's of the 80s) and the fact that
> >> by
> >> the 90s and certainly by now, what was relevant on the high end
> >> mainframes
> >> (such as the Cray or IBM 360/370's), were relevant to the home micro and
> >> dedicated units.

> > You sound just like the anti-thesis of Bob
> > Hyatt. Every time he was challenged to
> > "port" his chess algorithm to some slower
> > hardware, he insisted it could not be done;

> I've seen the printed source to a couple versions of CrayBlitz. (I still
> have a scanned copy, too.) (He keeps telling me that he'll release it
> publicly Real-Soon-Now, as soon as he can get the book to work. I think he
> will eventually publicly release it, it's just that he's in no hurry to do
> so. He has other things to do and it's just not that interesting for him.)

Oh, I'm sure he has been much too busy for
all of thirty years or so.


> He's right. It's not reasonable to port CrayBlitz to another system.

When a person is "right", he does not ever
need to contradict himself the way BH did.

For instance, I noticed that in many of the
old discussions he wanted/needed to have
things both ways. If they were talking about
his various chess programs, BH might write
something to the effect that one was in
essence the same as another, only on
different hardware. But if someone else--
say, one of those who wanted to know how
"smart" the program was vs. how "fast" was
a Cray-- would ask an annoying question,
Mr. Hyatt would simply contradict himself.
This lent the impression that he was not
merely wrong, but averting the truth.


> Many of the *ideas* are relevant to any system, but the implementation (and
> some of the algorithms too) was too tuned to the Cray to be suitable for any
> other system.

Indeed, that was not exactly the question he
averted. (When the creator of the Rybka engine
was recently asked about some problems in
its endgame, he simply commented that he
had not devoted much time/effort to that area,
thinking it would not pay off. No ego crumpling,
no offense taken, nothing. Quite a contrast.)


> Porting a Cray tuned chess program to another system is not a trivial task.
> Porting a regular program to the Cray is also not a trivial task if you want
> to use the vector hardware.

Certainly, not one of those people ever
asserted that it was. My impression was
that they wanted to know how much and
how good was the chess knowledge of the
program, in contrast to how fast a Cray or
how many nodes it could examine.


> The reality is, the Cray architecture is highly unsuited to computer chess.
> Chess does not vectorize well, at all. That he managed to make it work as
> well as he did is a testiment to his & Nelson's programming skills.

Well, if the thing was just junk, they might
have crated it and mailed it to me. I would
gladly have sent them in return my old Boris
Diplomat, and Atari2600 with chess cartridge,
and all my baseball cards to boot. ; >D


> He's not the best chess programmer in the world. He's good at it, but he's
> not good at tuning things just right. And since he's been using such strong
> hardware since 1981, he's used to it and depends on it now.

I noticed that when he kept talking about
all the money he spent to keep near the
bleeding edge hardware-wise.


> Hyatt has definetly done a lot of solid research over the years. He's done
> more testing and experimentation than probably anybody else in the world.

He certainly seemed to enjoy operating
Crafty on ICC.


...memory test completed. All 48k are
good to go now. Load floppy disk and
prepare to stand by while searching for
DOS. Have error-code book handy...


-- help bot



   
Date: 07 Jul 2008 20:19:30
From: Guest
Subject: Re: When only 1 legal move, GetClub will not think more.
"help bot" <[email protected] > wrote in message
news:f54da0b7-d50a-489f-893c-6ae902f5d681@l42g2000hsc.googlegroups.com...
> On Jul 6, 11:52 am, "Guest" <[email protected]> wrote:
>
>> And there were a heck of a lot more chess programs for the 8 bit systems,
>> at
>> least until the PC really took off in the mid 80s.
>
> I don't recall having much in the way of choice,
> until I got a PC, that is. They had a cartridge--
> you could buy it, or not, your choice. :>D

Oh, there were lots of chess programs in the early 80s. I remember getting
computer magazines around 80 (at the earliest) and onward, and there were
quite a few.

I don't remember them all, but here are a few from that time period:

MyChess, MicroChess, Sargon 1 & 2, Sfinks (of course), Micro-Chess (not
Peter Jennings), Video Chess, 8080 Chess. Chess 0.5 (Pascal source, not
much of a chess program.)

And those are just off the top of my memory from the period 78-81ish. And
there were dozens more competing in tournaments. The odds are good at least
a few of them were sold commercially too (like Sargon & MyChess & Sfinks
were.)

There were even more in the 81-85 years when 8bit micros were still being
widely used.

My little unpopular micro (RadioShack Color Computer) had... ChessD,
MicroChess, Cyrus, VoxChess, Chess, and at least one more that I can't
remember the name of. And that was an unpopular system.

Something popular like the C64 or Apple ][ would have at least double that.




>> I've seen the printed source to a couple versions of CrayBlitz. (I still
>> have a scanned copy, too.) (He keeps telling me that he'll release it
>> publicly Real-Soon-Now, as soon as he can get the book to work. I think
>> he
>> will eventually publicly release it, it's just that he's in no hurry to
>> do
>> so. He has other things to do and it's just not that interesting for
>> him.)
>
> Oh, I'm sure he has been much too busy for
> all of thirty years or so.

Actually, that's not quite the problem. At least not for the past 15 years
or so. (Or least since shortly after he stopped CrayBlitz and started work
on Crafty. Back in the days of CrayBlitz, he certainly would not have
released it.)

For at least the past 10 years, he didn't *have* a copy. It was gone. (And
no, that is *not* unusual. You'd be surprised how many chess programmers
have told me they no longer have a working or readable copy of their old
program. With disk crashes, lost backups, lost tapes, unreadable tapes,
fires, and so on, it's not at all unusual.)

The electrnoic copies he had were for tape readers he no longer had access
to and they had degraded beyond readability anyway. (When you can see
through a tape, you can be pretty sure it's not readable anymore...) One
tape that he thought was still good had read errors. (This was years ago
when I asked about CrayBlitz.)

It was only at my prodding a few years ago that he went through his library
and found two printed copies. (You had to take a printed copy of your
program to tournaments, so if there were any acusations of clone or theft,
etc., you could show the source.)

He even called Harry Nelson and asked him to go to Cray and somewhere else
(can't remember the name...) and see if they still had a copy of CrayBlitz.
They didn't.

I scanned them in the printed copies and tried to OCR them, but without much
luck. (Have you ever tried to OCR Fortran??? (shudder))

It was nearly a year later than he managed to find somebody who had a
*partial* electronic copy. The opening book stuff was gone, so he will have
to cobble something together from the printed stuff, if it's similar. Plus
he used several Cray specific fortran commands that GNU Fortran apparently
doesn't like. Meaning it wont run.

He's either going to have to rewrite the book stuff or release it without
any book at all. Neither of which is a fun choice.


>> He's right. It's not reasonable to port CrayBlitz to another system.
>
> When a person is "right", he does not ever
> need to contradict himself the way BH did.
>
> For instance, I noticed that in many of the
> old discussions he wanted/needed to have
> things both ways. If they were talking about
> his various chess programs, BH might write
> something to the effect that one was in
> essence the same as another, only on
> different hardware. But if someone else--
> say, one of those who wanted to know how
> "smart" the program was vs. how "fast" was
> a Cray-- would ask an annoying question,
> Mr. Hyatt would simply contradict himself.
> This lent the impression that he was not
> merely wrong, but averting the truth.


Not quite sure what you are saying there.

The program itself had some chess knowledge. Quite a bit. (I didn't
examine the algorithms that closely, so I can't say what.)

Except for the parallel processing stuff, there probably wasn't much that
was truely unique or unusual. But what was there was very often very Cray
specific. Things were often chosen and coded to make the Cray happy.

He did depend heavily on the Cray's raw performance to work as well as it
did.

It was kind of like how Ken Thompson gave up on regular computers and built
Belle. Hyatt gave up on regular computers and used a Cray.

(In fairness, Hyatt wasn't the only one who used a super computer. Hans
Berliner was the first one when he used an IBM 360/91 for J.biit. Slate &
Atkin used a super computer for Chess 4.x Slate later used a Cray for
NuChess. And so on.)


>> Many of the *ideas* are relevant to any system, but the implementation
>> (and
>> some of the algorithms too) was too tuned to the Cray to be suitable for
>> any
>> other system.
>
> Indeed, that was not exactly the question he
> averted. (When the creator of the Rybka engine
> was recently asked about some problems in
> its endgame, he simply commented that he
> had not devoted much time/effort to that area,
> thinking it would not pay off. No ego crumpling,
> no offense taken, nothing. Quite a contrast.)


>> Porting a Cray tuned chess program to another system is not a trivial
>> task.
>> Porting a regular program to the Cray is also not a trivial task if you
>> want
>> to use the vector hardware.
>
> Certainly, not one of those people ever
> asserted that it was. My impression was
> that they wanted to know how much and
> how good was the chess knowledge of the
> program, in contrast to how fast a Cray or
> how many nodes it could examine.

Okay. Now I understand.

And that's a question you can't answer without a lot of work.

You can't easily seperate the Cray out of CrayBlitz. It's not called "Blitz
on Cray".

It would take a massive rewrite to get the Cray out of it and make it
suitable for a regular computer. Not just a translation, but you'd have to
rewrite everything that depended on the Cray hardware.

Attack generation. Move generation. Evaluator. Parallel processing.
Trans table.

All total, v47a is about 16500 lines and 800k, excluding the CAL, which is
about another 300k and 9000 lines of Cray Assembler Language source.


Realisitcally I don't know how much would have to be modified. But I would
guess perhaps a quarter would have to be rewritten, and another quarter
examined and slightly modified for a different hardware and OS. (And not
just simply 'rewritten' but actually some effort expended to rewrite it well
and tune it to the new hardware.)

He did have a portable version, but it was basically just the regular
Fortran version doing things sequentially. It was still set up for vectors
and so on, it was just done sequentially on a regular computer. And as a
result, it ran very slowly.

Figure a reduction by 10 for regular hardware. Probably not unreasonable.
That would cut about two plies off its search. The reduction in hash table
size would hurt a little too, but probably not a lot.

Figure at least 100 points ratings drop, but I can't say exactly of course.



>> The reality is, the Cray architecture is highly unsuited to computer
>> chess.
>> Chess does not vectorize well, at all. That he managed to make it work
>> as
>> well as he did is a testiment to his & Nelson's programming skills.
>
> Well, if the thing was just junk, they might
> have crated it and mailed it to me. I would
> gladly have sent them in return my old Boris
> Diplomat, and Atari2600 with chess cartridge,
> and all my baseball cards to boot. ;>D

I didn't say it was junk.

I said it was highly unsuited for computer chess.

Chess doesn't vectorize well. Even trying to do chess with today's SSE
stuff is a pain. Cray's vectors would have been worse.

A lot of computer problems don't vectorize well.


>> He's not the best chess programmer in the world. He's good at it, but
>> he's
>> not good at tuning things just right. And since he's been using such
>> strong
>> hardware since 1981, he's used to it and depends on it now.
>
> I noticed that when he kept talking about
> all the money he spent to keep near the
> bleeding edge hardware-wise.

He hasn't spent as much as he would have if Cray, AMD, Intel, and his
University hadn't donated lots of computer time and hardware...

Considering the cost of Cray time, he would have needed to sell his house to
pay for *one* tournament.


>
>
>> Hyatt has definetly done a lot of solid research over the years. He's
>> done
>> more testing and experimentation than probably anybody else in the world.
>
> He certainly seemed to enjoy operating
> Crafty on ICC.

I think he does.

In the old days, it wasn't easy to do a chess program.

You could make it available on the time share systems and get some feedback
that way.

You could enter in tournaments and actually go there and play a few games.

But it was slow and tedious and required quite a bit of effort on your part.

But now, you set the program up on a spare system, let it run all by itself,
and you can play lots of games and do lots of testing with practically no
effort on your part.

(Of course, he does a lot of his testing on the server farms that his
University has. He probably has access to 1000 cores (total), so he can run
several hundred games at once and get reliable test results within hours,
where as most people would take days or weeks.








----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
---= - Total Privacy via Encryption =---


  
Date: 06 Jul 2008 09:18:29
From: Sanny
Subject: Re: When only 1 legal move, GetClub will not think more.
> Play a serious match (20 games at least) at regular time controls against
> some of the old programs or against some dedicated units.
>
> His practice of playing one game against Rybka at different time controls
> just doesn't work. It's mathematically unsound. One game just is not
> enough. At least 20 games at comparable time controls are needed for
> reasonable reliability.

Rybka is world best So I play against it. I find it interesting that
how number of moves taken by Rybka is increasing every week.

3 months back Rybka used to win Easy Level in 25 moves.
2 months back it started taking 30 moves / average.to win
1 month back it started taking 35 moves / average to win
Now it takes 40+ moves to win the Easy Level.

Play Chess at: http://www.GetClub.com/Chess.html

So that gives me satisfaction that game is really improving.

Help Bot has not played games with Easy Level yet. When he will play,
he will tell you about the improvements he finds in GetClubs gameplay.

It is very enjoying to play at GetClub. You can play upto 120 Games
for free. And win prizes every month So you can get 100-200 more games
for free easily.

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html


   
Date: 06 Jul 2008 12:42:14
From: Guest
Subject: Re: When only 1 legal move, GetClub will not think more.
"Sanny" <[email protected] > wrote in message
news:84efb8fb-0afa-4ffe-a20e-589262ff13eb@k37g2000hsf.googlegroups.com...
>> Play a serious match (20 games at least) at regular time controls against
>> some of the old programs or against some dedicated units.
>>
>> His practice of playing one game against Rybka at different time controls
>> just doesn't work. It's mathematically unsound. One game just is not
>> enough. At least 20 games at comparable time controls are needed for
>> reasonable reliability.
>
> Rybka is world best So I play against it. I find it interesting that
> how number of moves taken by Rybka is increasing every week.

But it's not representative.

Your testing method is flawed. Seriously flawed.

First, you are tuning your program to play gainst Rybka. That says very
little about how it would play against others or people. Yes, you are
taking some advice from people who play, but by doing most of your private
testing with Rybka, you are tuning to play against Rybka.

Let's pretend that Rybka has a major weakness... Let's say that it always
weakens the king side. By concentrating your testing with just Rybka, you
would be focusing on the expected weakness. But what would your program do
with an opponent who didn't have that weakness?

Why do you think people keep telling you to do your testing against a wide
range of other programs? GNU Chess, MicroMax, Crafty, Fruit, and so on?

By concentrating on one program to do most of your tuning, you are basically
teaching it to play against Rybka. How do you think your program is going
to perform if you play against somebody who has very different style?

I'm reminded of David Levy's old advice "Doing nothing, but doing it well"
for playing against programs. That would be a very very different style
than what Rybka plays. How will your program cope?



> 3 months back Rybka used to win Easy Level in 25 moves.
> 2 months back it started taking 30 moves / average.to win
> 1 month back it started taking 35 moves / average to win
> Now it takes 40+ moves to win the Easy Level.

All that says is that you are slowly making your program less terrible. Or
that it's faster and just isn't making as many tactical blunders.

That doesn't say anything how well it really plays real games.

And by doing Rybka at a major time handicap, you are preventing it from
performing as well as it can. Do you really think Rybka's play at 5 seconds
a move (or whatever you are doing this week) is truely representative of how
Rybka plays? At that quick time control, it's mostly just tactical play.

And it saying that it takes 40+ moves to loose means nothing. A loss is
still a loss. Let me repeat that... a loss is still a loss. All it means
is that now your program has learned how to delay it. That's not the same
thing as winning. If you played in a real tournament and you lost in 50
moves, do you think the tournament organizers are going to give you a point
for taking 50 moves to LOOSE?


And you can *NOT* just do one or two games and assume that your program is
better now than it was before. Testing takes at least a couple dozen games.
Preferably HUNDREDS to be able to get a wide range of games. With either
random books or some sort of even randomness so the programs wont play the
same sequence of moves over & over.

Just sheer random chance can result in a program (or person) making a bad
move in a game. If you do your quick test match and that bad luck shows up,
you can throw out a very good idea as being bad, all because of bad luck.

For example, say you introduced a bug that prevents it from being able to
win in an endgame where you have a queen & rook against a lone king. Unless
you play enough games to encounter this kind of position, you'll never know
about it. (Yes yes yes... I know this is a stupid example. I'm just trying
to simplify it enough to make the point obvious.)



> Play Chess at: http://www.GetClub.com/Chess.html
>
> So that gives me satisfaction that game is really improving.
>
> Help Bot has not played games with Easy Level yet. When he will play,
> he will tell you about the improvements he finds in GetClubs gameplay.

Unless he plays at least a couple dozen games, he can't truthfully say
whether it has improved or not.

Just playing one or two games and saying it's worse or better is inaccurate.
That's what I was telling you before.

The interactions between the chess heuristics and the search algorithms can
cause all sorts of unexpected side effects. Weaker play in some areas but
stronger play in others.

Lets say you improve your pawn structure, but in doing so, you cause the
program to like 'bad' bishops. Unless he happens to play a game where the
'bad' bishops are a factor, he wouldn't even notice the problem.

Unless you do enough games to get a good spread, you can't reasonably say
whether a change is good or bad as a whole.


I would say at least 20 games against several opponents, and preferably a
hundred games or more against at least 4 different programs, including your
previous program. And make sure you have some randomness somewhere so it
wont play the same opening lines over & over.

Automate it so the tests can be done without human participation, and just
let it play for a couple days or a week, and then decide whether the change
is good or bad.

(This is why serious chess programmers have computer 'farms' with at least 3
or 4 identical systems. And the commercial people will likely have dozens
of identical systems to do their testing on.)


Alternatively, if you don't want to go that route, modify the program to
play on the various free chess servers. That way you can get a much wider
range of players and actually get a rating as well.



Yes, I admit it, doing testing is a pain in the ass. It's real easy (and
tempting) to play just a game or two and declare a change as being better or
worse. It's very easy & tempting.

But it's wrong. If you are going to be serious about a chess program, then
you have to be serious about the testing. If not, then you wont much good
progress and there's no reason for anybody to take you seriously.



> It is very enjoying to play at GetClub. You can play upto 120 Games
> for free. And win prizes every month So you can get 100-200 more games
> for free easily.

No thank you. I have no desire to play your program. I have enough
programs and chess machines as is. I don't feel the urge to spend my spare
time playing your program.

But thank you for the offer.



> Bye
> Sanny
>
> Play Chess at: http://www.GetClub.com/Chess.html




----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
---= - Total Privacy via Encryption =---


  
Date: 05 Jul 2008 19:25:04
From: help bot
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 5, 11:53 am, "Guest" <[email protected] > wrote:

> If it's handy, I'd like to see the info.

Way back then, they had a magazine called
"Computer Chess Reports" or something like
that. They wanted "big money" for their
research, which explains why I only got my
hands on a very few issues. A complete
collection would reveal a much better picture.


> It wouldn't be relevant to the micro's at the time. (Other than Nelson
> showing that even small tables could benefit micro's of the time.) Relevant
> for the micro's of the 90s, but not the 80s.

The 1990s... I am thinking of the Fidelity
Par Excellence, the Novag Constellation
and Super-Constellation, and on up well
into USCF Master territory. All of these
would simply blow away Sanny's program,
Pentium or no Pentium. I never had the
best models, because they were too
expensive and constantly rendered no
longer cutting edge by the mere passing
of time.


> Their opening book was probably hard wired as randomly picking one of two or
> three moves.

The one I described as having no book
had no book-- as in none. It would think
about its first move for a long time, then
perhaps play 1. Na3, because it didn't
hang any material.


> > As I recall, my first computer which had sufficient
> > memory for hash-tables (as described in the earlier
> > post) was a PC, but it wasn't until I played Now Chess

> Right.... That's one of the reasons I've been wondering how many of the
> early PC or home micro programs used hash tables. 64k or above would have
> been enough for at least some hash tables, if not a full transposition
> table.

Now Chess was not considered to be
among the top programs, but it simply
"dismantled" my Colle System, like a
wrecking ball.

You keep mentioning 64k, which of
course was the total amount of memory
in the Commodore64 computer, which
dramatically undercut Atari on price.
But it was ultimately the PC that won
out, because there's more to life than
just playing games.


> Heck, the hardware based Belle chess computer searched 110k positions per
> second and had a rating of 2100+, but it used only 128k for transpositions.

In computer vs. computer events, speeding
up, say, a King and pawn ending, might be
decisive.


> So 64k, like on an Apple 2 or Commodore 64, would have been enough. So home
> micro's certainly had enough memory. It's just a question of whether the
> author wanted to or not. I have no information on when that happened.

I can't remember. But I do recall the "experts"
of that time talking about how it wasn't enough
yet, and then prices kept on dropping and the
additional memory became "standard" on PCs.


> The only reasons I've mentioned it are because of the hash table test Nelson
> did (which is very relevant to the micro's of the 80s) and the fact that by
> the 90s and certainly by now, what was relevant on the high end mainframes
> (such as the Cray or IBM 360/370's), were relevant to the home micro and
> dedicated units.

You sound just like the anti-thesis of Bob
Hyatt. Every time he was challenged to
"port" his chess algorithm to some slower
hardware, he insisted it could not be done;
that it would be grossly unfair and irrelevant
because he designed it to be hardware-
specific. Essentially, the critics wanted to
know why the Cray-beast was not even
better than it was (even if it was the best),
since that machine was a tad quicker than
your average turbocharged Chevy V8. : >D


> When you think about it... the 8 bit micro's of the early 80's had just as
> much ability as those old PDP's.

Explain THAT to the college, which presumably
paid tens of thousands of dollars "for nothing". ; >D


> You have to keep things in perspective. The old systems of the 60's & 70s
> were easily outclassed by what the micros of the 80s could do.
>
> So a micro chess programmer of the early-mid 80's had as much computing
> power and nearly as memory, as the systems where the early chess programs
> were developed.
>
> Yes, it still depends on whether the programmers chose to do those things.

The reality is that each of the programmers
or teams wrote programs for various different
machines; things were in a constant state of
flux, and it wasn't clear who would ultimately
be the "winner" of the hardware wars. Those
who stuck with Atari, Commodore or Apple
lost out, while those who chose to focus on
the PC early-on benefited.


> I can't say whether they did or not. Perhaps you can find your old
> ChessLife's and see. That would answer some of these questions once & for
> all.

The big advertisers like Fidelity and Mephisto
would reveal their state-of-the-art in these old
magazines, but not all of them were so quick
to spend the big bucks to promote their
programs. For instance, I found out about a
program called "Mchess" by word of mouth!


> > be a sort of /theoretical/ vs. reality rift, where on
> > one side we have an analysis of what mainstream
> > chess programmers /ought/ to have tried, and on
> > the other we have me, relating (though it is all a
> > bit foggy now) how things actually were in the
> > trenches, armed only with our M1 rifles... .

> It's not about "ought" to have.

> But at what point were the micro systems capable of doing it. That has to
> come first.

What I said. You are talking about the
theory and the theoretical possibilities,
while what I was originally describing were
the realities, what in fact happened to us
consumers in the real world.

The "experts" in many cases were just
plain wrong, and so it comes as no surprise
(to me anyway) to find that a sort of 20/20
post mortem which concludes that they
both couldda and shouldda used such
things earlier than they did. It is unfortunate
that much of the huffing and puffing by old
time "experts" cannot easily be retrieved;
that it was not done right here on this forum.
I think it would be very interesting to do a
comparison of what was once considered
"expert opinion" and what we know now,
with 20/20 hindsight.


> I took a quick browse through some of my stuff, and the earliest home micro
> based chess program that used transposition tables was "Sfinks X". Sfinks X
> was simply the experimental tournament version 'Sfinks"
>
> Sfinks was written by William Fink, and it ran on a TRS-80 Model 1 & 3 (with
> a z-80). This was about 1982 or so. It used about 30k program and 14k
> data.

Radio Shack is in fact where I purchased
my first "computer"... but not the TRS-80.

I have never even heard of that program
until now-- "Fink" being either a rat, or else
the guy whom Disney named some boat ride
after (i.e. Mike Fink).



> That's a bit earlier than what I was expecting, but that is what the
> reference says. Another one says that Sfinkx 3 (which ran on the PC) came
> out in 1983 and it used refutation tables. It doesn't mention transposition
> tables. It used 18k program and 20k data.

I had a PC, and I had never heard of Sfinkx.
How about something like "Sargon"?


> I don't know about any earlier home micro based programs that used a
> transposition table.
>
> I have no technical info on the early dedicated units, so I don't know when
> they did trans tables either.

Me neither. We would need to see an
archive of all the CCR magazines, or
perhaps probe online newsgroups that
preceded this one. I believe it sort of
happened all by itself, as more and more
memory became "standard" on PCs, and
programmers just decided to give it a try.

That was then, this is now. Now, rich
folks can go to Best Buy and gobble up a
humunga hard drive, storing endgame
table-bases. Now, the strongest chess
computer in the world is your own PC--
running download-able software named
after a fish.


-- help bot





   
Date: 06 Jul 2008 10:52:54
From: Guest
Subject: Re: When only 1 legal move, GetClub will not think more.

"help bot" <[email protected] > wrote in message
news:54140ff3-d532-472e-9aa6-2a696934a7bb@r66g2000hsg.googlegroups.com...
> On Jul 5, 11:53 am, "Guest" <[email protected]> wrote:


>> It wouldn't be relevant to the micro's at the time. (Other than Nelson
>> showing that even small tables could benefit micro's of the time.)
>> Relevant
>> for the micro's of the 90s, but not the 80s.
>
> The 1990s... I am thinking of the Fidelity
> Par Excellence, the Novag Constellation
> and Super-Constellation, and on up well
> into USCF Master territory. All of these
> would simply blow away Sanny's program,
> Pentium or no Pentium. I never had the
> best models, because they were too
> expensive and constantly rendered no
> longer cutting edge by the mere passing
> of time.

My first one was the mini sensory chess challenger, then the Par Excellence.
Later I bought a little novag portable that I can't remember the name of.
Plus I bought several chess programs and 'obtained' a few more. Right now,
I have a couple portables, including a little keychain chess computer. (Not
too strong, but cute. Perfect for playing chess while driving down the
freeway....(grin))

A guy I know still has the Par Excellence and the super connie and he used
to like to do matches against some of the freeware programs, like TSCP, etc.

Never having played GetClub chess (and I have no desire to) I can't really
comment about its strength.

But I admit I am curious about how it would compare to the older programs
(sargon 2) or older harrdware.

Play a serious match (20 games at least) at regular time controls against
some of the old programs or against some dedicated units.

His practice of playing one game against Rybka at different time controls
just doesn't work. It's mathematically unsound. One game just is not
enough. At least 20 games at comparable time controls are needed for
reasonable reliability.

I have suggested to him that he actually do a few matches against some of
the older programs or even those such as GNU Chess, micro-max, etc.
Obviously it hasn't happened.



>> Their opening book was probably hard wired as randomly picking one of two
>> or
>> three moves.
>
> The one I described as having no book
> had no book-- as in none. It would think
> about its first move for a long time, then
> perhaps play 1. Na3, because it didn't
> hang any material.

That's really odd.

It takes just a few bytes of code to hardwire a couple moves into it.

Even Sargon 1 did that, and it was definetly not a sophisticated program.

You must be talking about one of the first Chess Challengers in the 70's or
something. Some of those were pretty limited hardware. Sometimes as little
as 2k of ROM.


>> > As I recall, my first computer which had sufficient
>> > memory for hash-tables (as described in the earlier
>> > post) was a PC, but it wasn't until I played Now Chess
>
>> Right.... That's one of the reasons I've been wondering how many of the
>> early PC or home micro programs used hash tables. 64k or above would
>> have
>> been enough for at least some hash tables, if not a full transposition
>> table.
>
> Now Chess was not considered to be
> among the top programs, but it simply
> "dismantled" my Colle System, like a
> wrecking ball.
>
> You keep mentioning 64k, which of
> course was the total amount of memory
> in the Commodore64 computer, which
> dramatically undercut Atari on price.
> But it was ultimately the PC that won
> out, because there's more to life than
> just playing games.

Ultimately, yes.

But I mention the 64k of memory because the Atari's and Apple ][ and
Commodore 64 and several other 8 bit micros had. (Or 48k plus 16k rom.)

And there were a heck of a lot more chess programs for the 8 bit systems, at
least until the PC really took off in the mid 80s.

There were probably at least 5 or 6 chess programs for every 8 bit micro.
There was simply more variety.

And most of the dedicated units were basically dedicated 8 bit micros, so
comparisons to the 8 bit home computers are somewhat reasonable.



>> Heck, the hardware based Belle chess computer searched 110k positions per
>> second and had a rating of 2100+, but it used only 128k for
>> transpositions.
>
> In computer vs. computer events, speeding
> up, say, a King and pawn ending, might be
> decisive.

Oh undoubtably.

For an endgame, a trans table is a major benefit. Even a small one can be a
massive help.

It's definetly one area where a refutation table isn't nearly as good.

But it's questionable wehther it would be worth spending another $50 or so
for 32k-64k of ram to put into an early 80's dedicated machine.

For a dedicated system, not only do you have to balance the benefits vs.
program cost, you also have to factor in the parts cost and whether that
will be offset by higher prices and the change in purchases.

Personally, I doubt they would consider it worth it unless the machine cost
us$500 anyway. At least not until the memory prices dropped enough for the
cost to be neglible.

There for quite a while, micro's were right at the edge of whether or not a
trans table would be much use in typical positions. Smaller hash tables
(for pawn & king evals) and refutation tables would probably be a better
choice for a lot less memory & hardware cost.

Home systems had an advantage in that area because the hardware was already
bought, so the program could use it without any extra cost.




>> So 64k, like on an Apple 2 or Commodore 64, would have been enough. So
>> home
>> micro's certainly had enough memory. It's just a question of whether the
>> author wanted to or not. I have no information on when that happened.
>
> I can't remember. But I do recall the "experts"
> of that time talking about how it wasn't enough
> yet, and then prices kept on dropping and the
> additional memory became "standard" on PCs.

I really wish I could find some old articles on experiments for trans table
sizes. It's probably a pdf I stored no telling where...

The smaller sizes are not a major benefit on the systems the 'experts' were
using. (Although Nelson did show that even for CrayBlitz, using the memory
for pawn & king structure hashes was a benefit, provided your code was
complicated enough to be worth the effort.)

The one area that I'm not so sure of is...

For the size trees the micro's were searching, would it have actually been
worth it.

The home micro's (8 bit) were searching probably 5-6 plies middle game. At
that depth, refutation tables are supposed to work just as well, but with
much less memory. (Width * Depth / 2 entries)

So it is a fuzzy area.

So it's possible it's not much of a benefit until you search fast enough to
go beyond what refutation tables can give you, or your program is
sophisticated enough to need to store pawn & king structure info into a
hash.

So you might need either faster processors or more advanced search
techniques (null move, etc.) to make better use of the limited hardware.
Then when you can reach 6+ plies in the middle game, trans tables might be
better.



>> The only reasons I've mentioned it are because of the hash table test
>> Nelson
>> did (which is very relevant to the micro's of the 80s) and the fact that
>> by
>> the 90s and certainly by now, what was relevant on the high end
>> mainframes
>> (such as the Cray or IBM 360/370's), were relevant to the home micro and
>> dedicated units.
>
> You sound just like the anti-thesis of Bob
> Hyatt. Every time he was challenged to
> "port" his chess algorithm to some slower
> hardware, he insisted it could not be done;

I've seen the printed source to a couple versions of CrayBlitz. (I still
have a scanned copy, too.) (He keeps telling me that he'll release it
publicly Real-Soon-Now, as soon as he can get the book to work. I think he
will eventually publicly release it, it's just that he's in no hurry to do
so. He has other things to do and it's just not that interesting for him.)

He's right. It's not reasonable to port CrayBlitz to another system.

It was so heavily vectorized that even modern 64 bit systems aren't
suitable.

Much of his program was designed specifically for a vector computer. Doing
things sequentially would work, but it would result in a major performance
hit. Plus you'd have massive cache issues and memory latency issues on
modern hardware.

(He had a portable test version, of course, that he used on a VAX or
whatever system was handy, but he said it suffered major performance issues
due to lack of vector hardware. I believe it.)

Many of the *ideas* are relevant to any system, but the implementation (and
some of the algorithms too) was too tuned to the Cray to be suitable for any
other system.

Porting a Cray tuned chess program to another system is not a trivial task.
Porting a regular program to the Cray is also not a trivial task if you want
to use the vector hardware.

The reality is, the Cray architecture is highly unsuited to computer chess.
Chess does not vectorize well, at all. That he managed to make it work as
well as he did is a testiment to his & Nelson's programming skills.


> that it would be grossly unfair and irrelevant
> because he designed it to be hardware-
> specific. Essentially, the critics wanted to
> know why the Cray-beast was not even
> better than it was (even if it was the best),
> since that machine was a tad quicker than
> your average turbocharged Chevy V8. :>D

Well... I guess it could be the same reason why Crafty isn't better than it
is, in spite of him having massive computer clusters to use for
tournaments....

He's not the best chess programmer in the world. He's good at it, but he's
not good at tuning things just right. And since he's been using such strong
hardware since 1981, he's used to it and depends on it now.

Hyatt has definetly done a lot of solid research over the years. He's done
more testing and experimentation than probably anybody else in the world.
He knows more about computer chess than almost anyone else.

But he's just not the best.

And then throw in the odd architecture of the Cray... A vector computer
doesn't really fit computer chess.



>> When you think about it... the 8 bit micro's of the early 80's had just
>> as
>> much ability as those old PDP's.
>
> Explain THAT to the college, which presumably
> paid tens of thousands of dollars "for nothing". ;>D

They were buying name and peripherals and support and a multi-user OS.

But for a single user, by the early to mid 80s, many micros were a good
match for the PDP's from the 60's and mid 70's. This is especially true if
you were talking about a 68k system instead of an 8 bit micro.

If you look at the specs, many of the PDP's instructions per second weren't
that spectacular. Their advantages were the peripherals (disk drives,
etc.), larger memories (which often couldn't be accessed by a user) and the
larger word sizes (which aren't an issue for chess programming).

And, of course, the VAX ws introduced in 77 and was replacing the PDP which
was outdated.


>> You have to keep things in perspective. The old systems of the 60's &
>> 70s
>> were easily outclassed by what the micros of the 80s could do.
>>
>> So a micro chess programmer of the early-mid 80's had as much computing
>> power and nearly as memory, as the systems where the early chess programs
>> were developed.
>>
>> Yes, it still depends on whether the programmers chose to do those
>> things.
>
> The reality is that each of the programmers
> or teams wrote programs for various different
> machines; things were in a constant state of
> flux, and it wasn't clear who would ultimately
> be the "winner" of the hardware wars. Those
> who stuck with Atari, Commodore or Apple
> lost out, while those who chose to focus on
> the PC early-on benefited.

Not until the PC got to at least the 286. Or at least the 12mhz 8088s.

The 8088 & 8086 were so slow that my old 8 bit micro was often as fast.

And the very fact that there were so many 8 bit micro's gives then an
advantage in this case... Namely that they probably did indeed do hash
tables & transposition tables long before the dedicated units or the IBM PC
systems.


>
>
>> I can't say whether they did or not. Perhaps you can find your old
>> ChessLife's and see. That would answer some of these questions once &
>> for
>> all.
>
> The big advertisers like Fidelity and Mephisto
> would reveal their state-of-the-art in these old
> magazines, but not all of them were so quick
> to spend the big bucks to promote their
> programs. For instance, I found out about a
> program called "Mchess" by word of mouth!



>
>
>> > be a sort of /theoretical/ vs. reality rift, where on
>> > one side we have an analysis of what mainstream
>> > chess programmers /ought/ to have tried, and on
>> > the other we have me, relating (though it is all a
>> > bit foggy now) how things actually were in the
>> > trenches, armed only with our M1 rifles... .
>
>> It's not about "ought" to have.
>
>> But at what point were the micro systems capable of doing it. That has
>> to
>> come first.
>
> What I said. You are talking about the
> theory and the theoretical possibilities,
> while what I was originally describing were
> the realities, what in fact happened to us
> consumers in the real world.
>
> The "experts" in many cases were just
> plain wrong, and so it comes as no surprise
> (to me anyway) to find that a sort of 20/20
> post mortem which concludes that they
> both couldda and shouldda used such
> things earlier than they did. It is unfortunate
> that much of the huffing and puffing by old
> time "experts" cannot easily be retrieved;
> that it was not done right here on this forum.
> I think it would be very interesting to do a
> comparison of what was once considered
> "expert opinion" and what we know now,
> with 20/20 hindsight.
>
>
>


>> I took a quick browse through some of my stuff, and the earliest home
>> micro
>> based chess program that used transposition tables was "Sfinks X".
>> Sfinks X
>> was simply the experimental tournament version 'Sfinks"
>>
>> Sfinks was written by William Fink, and it ran on a TRS-80 Model 1 & 3
>> (with
>> a z-80). This was about 1982 or so. It used about 30k program and 14k
>> data.
>
> Radio Shack is in fact where I purchased
> my first "computer"... but not the TRS-80.

Me too.


> I have never even heard of that program
> until now-- "Fink" being either a rat, or else
> the guy whom Disney named some boat ride
> after (i.e. Mike Fink).

I vaguely remember Sfinks, because of the odd name, but it didn't really
mean anything to me back then. I don't know anything beyond what little I
read in a few reference books. (Welsh's two "Computer Chess" books. are the
only things I have that describe any details about it.)

I think there was also a chess computer company called Sphinx. Totally
unrelated to the program in spite of the similar sounding name.


Sfinks played in a few tournaments, such as the ACM games, the 81 micro
tournament in Paris, and the 83 world championship. I'm not sure when it
finally disapeared.

v1.81 was first sold in 1981 and could play several chess variants too. No
idea if that used any hash tables or if that was added for v2.



>> That's a bit earlier than what I was expecting, but that is what the
>> reference says. Another one says that Sfinkx 3 (which ran on the PC)
>> came
>> out in 1983 and it used refutation tables. It doesn't mention
>> transposition
>> tables. It used 18k program and 20k data.
>
> I had a PC, and I had never heard of Sfinkx.
> How about something like "Sargon"?

Sargon is actually a difficult subject...

There were so many versions of Sargon.

Sargon 1 was written in 1978.

Sargon 2 was written by 1979. (Ken Thompson, of Belle, helped a bit with it
too. Some ideas, advice, etc.)

Sargon 2 was converted to 6502 & v2.5 was put onto several dedicated
systems.

Sargon 3 was written in 1979.

Seems pretty simple (for the limited history I mention), but Sargon 2
continued to be sold for years as if it was state of the art.

Sargon 2 went commercial in about 80, I think. So the commercial versions
might be similar to the tournament version.

Sargon 3 didn't go commercial until 1983, I think. Four years after it was
written and competed in tournaments.

So I'm not so sure that the commercial Sargon 2 & 3 for home micro's were
actually the same as the 1979 versions they competed with.

It's entirely possible the chess engine core had been modified to take
advantage of the extra memory on whatever system it was being sold for.

Meaning that Sargon 2 or 3 for the Vic-20 might not have been the same one
that ran on the Apple ][ or the PC or that played in the earlier
tournaments.

There are probably experts on Sargon, but I guess the best person to check
with would be the Spracklens themselves. But I don't know their email
address. Just an old postal address.


I wonder if anybody has disassembled Sargon 2. I remember back in 83, a guy
gave me a hex dump of Sargon 2 for the Vic-20 and I started doing a
disassembly of it, but I never got done.


As a side note, Sargon V supposedly had a rating of around 2276 on a PC/AT.
Not too shabby. Assuming it's accurate....!



>
>
>> I don't know about any earlier home micro based programs that used a
>> transposition table.
>>
>> I have no technical info on the early dedicated units, so I don't know
>> when
>> they did trans tables either.
>
> Me neither. We would need to see an
> archive of all the CCR magazines, or
> perhaps probe online newsgroups that
> preceded this one. I believe it sort of
> happened all by itself, as more and more
> memory became "standard" on PCs, and
> programmers just decided to give it a try.
>
> That was then, this is now. Now, rich
> folks can go to Best Buy and gobble up a
> humunga hard drive, storing endgame
> table-bases. Now, the strongest chess
> computer in the world is your own PC--
> running download-able software named
> after a fish.
>
>
> -- help bot
>
>
>
>







----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
---= - Total Privacy via Encryption =---


  
Date: 04 Jul 2008 21:31:02
From: help bot
Subject: Re: When only 1 legal move, GetClub will not think more.
I agree with all this "would have/could have" stuff, but
they just didn't. It's akin to saying that one man could
have dug a tunnel through a mountain -- if he had done
this or tried that -- but there was no tunnel actually dug,
so everybody had to climb the mountain.

I probably still have a few of the old computer chess
magazines somewhere, and I certainly have some old
Chess Life magazines which contained information
regarding who had what -- hash-tables and so forth --
if nobody else here can drudge up the relevant
information regarding the more affordable chess
programs (no Crays, please-- the mere thought of the
electricity bills are killing me!)

Some of the earliest contraptions I played (not super-
computers) had zip-- one even had no openings book!

As I recall, my first computer which had sufficient
memory for hash-tables (as described in the earlier
post) was a PC, but it wasn't until I played Now Chess
(which handily demolished my Colle System) and then
Psion, that I ran into serious difficulties. Psion
seemed to be strong in tactics... and fast. Still, I'm
not at all certain whether even their successors used
hash-tables much, but if they did it was probably very
limited due to the relatively high cost (for mere mortals)
of memory back then.

Many of the programs mentioned here by others
are not -- I say NOT -- the widely-available, mass-
marketed ones that ordinary folks were buying in
stores at the time. Certainly, the Cray was not to
be seen in any of my local Service Merchandise
or later, Best Buy stores. Nor Mack-the-Hack, nor
the PDP/11 my college leased, etc. It seems to
be a sort of /theoretical/ vs. reality rift, where on
one side we have an analysis of what mainstream
chess programmers /ought/ to have tried, and on
the other we have me, relating (though it is all a
bit foggy now) how things actually were in the
trenches, armed only with our M1 rifles... .


-- help bot




   
Date: 05 Jul 2008 10:53:13
From: Guest
Subject: Re: When only 1 legal move, GetClub will not think more.
"help bot" <[email protected] > wrote in message
news:e08160cd-8f61-4001-86a4-d5f47365a811@k30g2000hse.googlegroups.com...
> I agree with all this "would have/could have" stuff, but
> they just didn't. It's akin to saying that one man could
> have dug a tunnel through a mountain -- if he had done
> this or tried that -- but there was no tunnel actually dug,
> so everybody had to climb the mountain.
>
> I probably still have a few of the old computer chess
> magazines somewhere, and I certainly have some old
> Chess Life magazines which contained information
> regarding who had what -- hash-tables and so forth --
> if nobody else here can drudge up the relevant
> information regarding the more affordable chess

That would be interesting.

If it's handy, I'd like to see the info.


> programs (no Crays, please-- the mere thought of the
> electricity bills are killing me!)

It wouldn't be relevant to the micro's at the time. (Other than Nelson
showing that even small tables could benefit micro's of the time.) Relevant
for the micro's of the 90s, but not the 80s.

The Cray-1 was interesting. I have a tech report on its hardware. They
didn't use any form of LSI, VLSI, etc. chips. It was built using 16x4bit
register chips, 1k bit memory chips, and 5/4 NAND chips.

Nothing more advanced than that because this way they could control signal
propogation times (the board size was almost the same as the clock
wavelength, so standing waves were easily generated and difficult to avoid)
and so that all the gates would be terminated so the power supply would only
have a DC load applied to it.

> Some of the earliest contraptions I played (not super-
> computers) had zip-- one even had no openings book!

Well, no. They probably had to fit everything into 4k of ROM.

Their opening book was probably hard wired as randomly picking one of two or
three moves.


> As I recall, my first computer which had sufficient
> memory for hash-tables (as described in the earlier
> post) was a PC, but it wasn't until I played Now Chess

Right.... That's one of the reasons I've been wondering how many of the
early PC or home micro programs used hash tables. 64k or above would have
been enough for at least some hash tables, if not a full transposition
table.


> (which handily demolished my Colle System) and then
> Psion, that I ran into serious difficulties. Psion
> seemed to be strong in tactics... and fast. Still, I'm
> not at all certain whether even their successors used
> hash-tables much, but if they did it was probably very
> limited due to the relatively high cost (for mere mortals)
> of memory back then.

64k would have been enough for some benefits.

Heck, the hardware based Belle chess computer searched 110k positions per
second and had a rating of 2100+, but it used only 128k for transpositions.

So 64k, like on an Apple 2 or Commodore 64, would have been enough. So home
micro's certainly had enough memory. It's just a question of whether the
author wanted to or not. I have no information on when that happened.



> Many of the programs mentioned here by others
> are not -- I say NOT -- the widely-available, mass-
> marketed ones that ordinary folks were buying in
> stores at the time. Certainly, the Cray was not to

No, it wasn't.

The only reasons I've mentioned it are because of the hash table test Nelson
did (which is very relevant to the micro's of the 80s) and the fact that by
the 90s and certainly by now, what was relevant on the high end mainframes
(such as the Cray or IBM 360/370's), were relevant to the home micro and
dedicated units.


> be seen in any of my local Service Merchandise
> or later, Best Buy stores. Nor Mack-the-Hack, nor
> the PDP/11 my college leased, etc. It seems to

The PDP/6 that MacHack VI was originally done on could only do 225k
instructions per second. The later PDP/10 managed 400k instructions per
second.

When you think about it... the 8 bit micro's of the early 80's had just as
much ability as those old PDP's.

You have to keep things in perspective. The old systems of the 60's & 70s
were easily outclassed by what the micros of the 80s could do.

So a micro chess programmer of the early-mid 80's had as much computing
power and nearly as memory, as the systems where the early chess programs
were developed.

Yes, it still depends on whether the programmers chose to do those things.

I can't say whether they did or not. Perhaps you can find your old
ChessLife's and see. That would answer some of these questions once & for
all.



> be a sort of /theoretical/ vs. reality rift, where on
> one side we have an analysis of what mainstream
> chess programmers /ought/ to have tried, and on
> the other we have me, relating (though it is all a
> bit foggy now) how things actually were in the
> trenches, armed only with our M1 rifles... .

It's not about "ought" to have.

But at what point were the micro systems capable of doing it. That has to
come first.

The early micro dedicated systems certainly were not.

But many of the home based micro's were.



I took a quick browse through some of my stuff, and the earliest home micro
based chess program that used transposition tables was "Sfinks X". Sfinks X
was simply the experimental tournament version 'Sfinks"

Sfinks was written by William Fink, and it ran on a TRS-80 Model 1 & 3 (with
a z-80). This was about 1982 or so. It used about 30k program and 14k
data.

That's a bit earlier than what I was expecting, but that is what the
reference says. Another one says that Sfinkx 3 (which ran on the PC) came
out in 1983 and it used refutation tables. It doesn't mention transposition
tables. It used 18k program and 20k data.


I don't know about any earlier home micro based programs that used a
transposition table.

I have no technical info on the early dedicated units, so I don't know when
they did trans tables either.






----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
---= - Total Privacy via Encryption =---


  
Date: 04 Jul 2008 04:59:17
From: Sanny
Subject: All improvements as suggested done.
All improvements as you suggested were done.

The program is now 20% faster than it was 3 days back.

Earlier the program used to improve 100% faster in a few days but now
only 20% improvement was achieved.

With even 20% improvement you will face a strong opponent.

Help Bot will find Easy level Tougher than before. I am waiting for a
few games by help bot against easy levels to see if the improvements
are working correctly or not.

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html





  
Date: 03 Jul 2008 21:40:51
From: help bot
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 3, 11:00 pm, "Guest" <[email protected] > wrote:

> Hash tables were first used in Mac Hack VI back in 1966, when it was first
> being written. That was the first truely significant chess program, as well
> as the first one that used transposition tables.
>
> Most significant mainframe chess programs also did hash tables.
>
> Microcomputer programs weren't as eager to add them for some reason.


One of the problems we have been having seems
to be that while I have been describing chess-playing
computers/programs which were widely available for
purchase, everybody else is intensely focused on
mainframes-- even discussing such things as the
Cray super-computer. In particular, my references
to Fidelity Electronics make a sharp contrast to
those less-affordable alternatives.

When I first got involved with computer chess, not
one of the commercially-sold machines used hash
tables to my knowledge, or any of the fancier
pruning techniques which are the hallmark of top
chess programs these days. I believe it was the
cost of memory which was the primary factor here,
just as the cost of really fast microprocessors was
the reason the strongest machines were so
expensive relative to weaker models like the ones
I used to play. Anybody remember the days when
the stronger table-top models used to overheat,
due to manufacturers attempting to "push the
limits"?


-- help bot


   
Date: 04 Jul 2008 10:29:57
From: Guest
Subject: Re: When only 1 legal move, GetClub will not think more.

"help bot" <[email protected] > wrote in message
news:dadb0954-2474-4c7f-a648-452fea7c22b5@j22g2000hsf.googlegroups.com...
> On Jul 3, 11:00 pm, "Guest" <[email protected]> wrote:
>
>> Hash tables were first used in Mac Hack VI back in 1966, when it was
>> first
>> being written. That was the first truely significant chess program, as
>> well
>> as the first one that used transposition tables.
>>
>> Most significant mainframe chess programs also did hash tables.
>>
>> Microcomputer programs weren't as eager to add them for some reason.
>
>
> One of the problems we have been having seems
> to be that while I have been describing chess-playing
> computers/programs which were widely available for
> purchase, everybody else is intensely focused on
> mainframes-- even discussing such things as the

That's partially because the mainframes of back then (when the early
research was done) are now slower than the hand helds and laptops of today.

Even the mainframes (and PDP's) of the 70's were out often classed by the
micro's of the 80s.

The CDC 6600 that Chess 3.x & 4.x was written on was a super computer of the
time. But the micro's of the 80s could out perform it for chess searching.

The PDP that MacHack VI was written on was a joke in the 80s.

And you have to keep in mind that the 6502 was actually better suited for
Chess than many of the mainframes. The only advantage the mainframes had
were larger memories. (Well, 32 bits could help for bitboards, but many
didn't use bitboards. They used regular mailbox boards.)

You have to keep your perspective here...

The super computers used when the early chess stuff was done are now so slow
and limited that you would have trouble running Windows 3.1 on them.
Although I guess a Cray-1 would have enough power to run Windows 95.



So if you talk about the early research and the stuff used way back then,
then you are relevant to the micro's used in the mid to late 80's and the
90's and certainly now.



> Cray super-computer. In particular, my references

I mentioned Cray because it was actually relevant to the discussion of
micros.

Harry Nelson (I thought it was Hyatt that wrote the paper, but it wasn't)
was showing how even a super computer could still benefit from tables as
small as what would have been relevant to micro's of the time.

That even at the speeds it searched, just a few K bytes (K bytes, not
megabytes) of hash tables were still beneficial.

That was certainly relevant to even the 8 bit micro's used in dedicated
chess computers and home micro's (such as Apple ][) that programs such as
Sargon etc. ran on.

Most micro's could have afforded a few K-Bytes of memory for a pawn & king
table. Even many dedicated systems could have managed it with only a few
dollars increase in price. A $10 increase in the top end system would have
been irrelevant if it allowed for better pawn & king analysis.

Which is what that part of the discussion was about. Micro programs of the
80s and hash tables.




> to Fidelity Electronics make a sharp contrast to
> those less-affordable alternatives.

Again... my coments were relevant to micros.

As long as you had 4k bytes or so to spare, you could reasonably use a hash
table for limited purposes. (Pawn & king structure.)

Any micro (home or dedicated) with 64k or more could have used a hash table
for transpositions as well.


> When I first got involved with computer chess, not
> one of the commercially-sold machines used hash
> tables to my knowledge, or any of the fancier

That's one of the reasons I mentioned I didn't know when hash &
transposition tables started showing up in the micro based systems.

Home micro's certainly had the space. 64k for the Apple ][ or Commodore 64
would have been enough. The 256k or 512k or even 640k for the IBM PC would
have been very useful. But I don't know when they started adding hash
tables.

For dedicated systems... Again, I don't know, but I would suspect probably
not until the 32 bit cpu's started being used. The assumption being that if
you were using a 6502, then you were so concerned about hardware prices that
you wouldn't want to spend the extra money for 32k of RAM (or more if you
bank-switched). And that if you were spending for a 32 bit cpu, you would
also spend for some more memory the hash tables.


I think the Spracklens were a bit behind the technology curve, actually. It
wasn't so much that the micros couldn't do it, as they weren't comfortable
doing them. The lower end stuff couldn't benefit because they probably only
had a a couple K bytes of memory. But the upper end stuff should have been
able to. But I don't think they did.

That would actually be a good question to ask them. But I don't know their
email address, just Dan's postal address and they probably wouldn't bother
to actually write out a reply.



> pruning techniques which are the hallmark of top
> chess programs these days. I believe it was the
> cost of memory which was the primary factor here,

Certainly it was a factor. I still remember when 64k dropped below $64.

Certainly prior to that, it wouldn't have been realistic except for a top
dedicated unit.

The early dedicated systems wouldn't have been able to afford it.

For a home system though, where the extra memory was already available... I
don't see why they wouldn't have gone ahead and done it, unless they weren't
comfortable with it.

I have no idea when hash tables started being regularly used.

Somewhere or other, I have some old papers on some early trans table
research that talked about the cost of maintaining hash tables and their
effectiveness for small sizes. But I have no idea where it's at,
unfortunately.

It's possible that doing it on an 8 bit micro was just too inconvenient for
the limited benefit you would get for such a short search depth.

That's why I suggested that maybe they prefered refutation tables instead.
Less memory and they still guided the search about as well.


Maybe it wasn't until the processors got fast enough to search 6+ plies
regularly that hash tables wouldn't have been significantly useful.

Mainframe & PDP etc. programs did it because they already had the memory
available. So even though it didn't give them much of a boost, it didn't
cost them anything either.



> just as the cost of really fast microprocessors was
> the reason the strongest machines were so
> expensive relative to weaker models like the ones

The cost between an overlocked 6502 and a 68000 for example were not that
major.

Not when you throw in the extra costs such as development, manufacturing,
sales, support etc. In that case, the hardware cost was small enough that
it was quite reasonable in the higher end systems.

I don't happen to have a handy SSDF list from back then, but if I remember
right, most of the 68000 & 68020's weren't rated that much higher than the
6502 systems.

Let me go check a few SSDF's. I'm curious.

Okay, in 1988, most were 8 bit systems with a couple 68k systems of mediocre
performance, with the exception of "Excel. Mach IIC+ 68000" at 1982 and the
Mephisto stuff at 2000+

So by 1988, the 6502 systems were reaching their limit in performance &
clock rate.

By 1989, there were more 68k and 68020 systems, but even still, some 8 bit
systems still outperformed the 68k systems.

In 1987, the Mephisto's were top, but I suspect that was programming skill
and not hardware because Mephisto MM4 at 5mhz was just 20 points less than
Mephisto Amsterdam 68k at 12mhz. Below those were a couple Fidelity
Excellence 68k systems, but they were just a few points above a 5mhz 6502
Fideltiy.


Of course, none of that really means much. We'd need to actually do some
checking to figure out when hash tables started showing up in the dedicated
and the micro programs.


To sumerize my rather rambling comments....

1) Even though Nelson showed (with CrayBlitz) that just a few kBytes of hash
table would be beneficial for a micro, I doubt many bothered to.

2) I think dedicated units probably used refutation tables before hash
tables.

3) The lower end dedicated units probably didn't even use refutation tables
because of the cost.

4) The upper end dedicated units could have done it, but I don't know if
they did.

5) It probably wasn't until the 32 bit systems until dedicated units started
doing trans tables.

6) For home computer programs, I doubt many commercial programs used hash
tables at all because many were based on programs from dedicated units.
(Sargon 2 & 3, etc.)

7) For home computers, most had enough memory (64k or more) for it to be
worthwhile having trans tables. So if the programs didn't, it would
probably had been either laziness or lack of familarity on the programmer's
part. It would have been feasible and probably beneficial enough to be
worth the extra cost of maintaining the hash index and hash tables.





> I used to play. Anybody remember the days when
> the stronger table-top models used to overheat,
> due to manufacturers attempting to "push the
> limits"?

I never had a top machine, so I can't say that I do. I knew of it, but I
never experienced it.

I do know that the overclocked processors were indeed hand picked for
overclockability.

The ones used for tournaments were pushed so hard they would sometimes crash
in the middle of a game.

You can see some pictures of the tournament micro's of the 80s with big fans
built into the boxes to try and keep them cool enough to survive.



>
> -- help bot




----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
---= - Total Privacy via Encryption =---


  
Date: 03 Jul 2008 17:13:08
From: help bot
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 3, 12:54 pm, Patrick Volk <[email protected] > wrote:

> Subroutine is an interesting choice of term. So basic.

BASIC? Beginner's All-purpose Symbolic Instruction Code.
It was my third language-- after FORTRAN IV and RPG
(Before your time, sonny.)


> Hmm. I'm 41, been in the business since I've been 19, and know hash
> tables were in chess programs in the 80's.

The late '80s, perhaps. But that was hardly the
beginning. Can you remember -- were you even
involved then -- when chess programs had none?


> > Wrong. A pawn can have anywhere from
> >zero legal moves all the way up to twelve,
> >depending on the position of course.

> Any piece can have a minimum of zero moves. Oh, and I forgot the
> promotion.

That's why you'll never become world champ.
You remind me of my old pal, Mr. Bogolyubov... .


> A-B is selective search. Even the massively parallel solutions use
> some form of selective search, if for nothing else, to explore the
> best-looking lines first.
> What's your point?

I haven't got one. The Wizard said I must first
bring him the broom of the wicked witch of the
west-- than I can get it.


> >> In other words, it's sorting. If you want to tell me a decent chess
> >> program can be made without sorting

That was /your idea/, not mine.


> >Try this link:http://www.gorillatrade.com

> And the point is... what, why don't they just have a mutual fund?

Sorry-- my mistake. I had assumed they would
have a pic of the Monkeys' singer "Davy" on their
Web page, just as in all their ads. As they didn't,
you -- as silly as you are -- probably got confused
by the "gorilla" and "monkey" similarities, thinking
they were the same species. You are probably
too young to even remember them-- the Monkeys
I mean.

Anyway, your suggestion to speed up the search
was a good one, except that it is even more crucial
for him to fix the problem with quiessence-search
first. I am still beating up on his program in simple
tactical skirmishes-- which indicates to me that it
is not calculating them properly (whether looking
ahead deeply or not). Speeding things up won't
help here, any more than say, eating donuts more
quickly will help unclog one's arteries.


-- help bot


   
Date: 04 Jul 2008 13:41:57
From: Patrick Volk
Subject: Re: When only 1 legal move, GetClub will not think more.
On Thu, 3 Jul 2008 17:13:08 -0700 (PDT), help bot
<[email protected] > wrote:

>On Jul 3, 12:54 pm, Patrick Volk <[email protected]> wrote:
>
>> Subroutine is an interesting choice of term. So basic.
>
> BASIC? Beginner's All-purpose Symbolic Instruction Code.
>It was my third language-- after FORTRAN IV and RPG
>(Before your time, sonny.)

Good, you can look up acronyms. What, next you'll be telling me you
had a chess program on the analytical engine.

>
>
>> Hmm. I'm 41, been in the business since I've been 19, and know hash
>> tables were in chess programs in the 80's.
>
> The late '80s, perhaps. But that was hardly the
>beginning. Can you remember -- were you even
>involved then -- when chess programs had none?

No, the early 80's.

And I'll reiterate, UNIX used hash tables. So did the 1ESS. Hmm... who
were them guys? They did a chess program as well. I'd mention it, but
it's a pretty straight giveaway. Did pretty well in the '82 ACM
tourney.

>
>
>> > Wrong. A pawn can have anywhere from
>> >zero legal moves all the way up to twelve,
>> >depending on the position of course.
>
>> Any piece can have a minimum of zero moves. Oh, and I forgot the
>> promotion.
>
> That's why you'll never become world champ.
>You remind me of my old pal, Mr. Bogolyubov... .

Never aspired to be world champ. Personally, I consider it good
business when my name doesn't come up, because it generally means
there ain't cuss words behind it.

Again, do you have a point, other than contradiction? Anything
constructive to add perhaps?

>
>
>> A-B is selective search. Even the massively parallel solutions use
>> some form of selective search, if for nothing else, to explore the
>> best-looking lines first.
>> What's your point?
>
> I haven't got one. The Wizard said I must first
>bring him the broom of the wicked witch of the
>west-- than I can get it.
>
>
>> >> In other words, it's sorting. If you want to tell me a decent chess
>> >> program can be made without sorting
>
> That was /your idea/, not mine.

No, your idea. Your idea because you said it's possible to write a
program without using existing strategies.

>
>
>> >Try this link:http://www.gorillatrade.com
>
>> And the point is... what, why don't they just have a mutual fund?
>
> Sorry-- my mistake. I had assumed they would
>have a pic of the Monkeys' singer "Davy" on their
>Web page, just as in all their ads. As they didn't,
>you -- as silly as you are -- probably got confused
>by the "gorilla" and "monkey" similarities, thinking
>they were the same species. You are probably
>too young to even remember them-- the Monkeys
>I mean.

*sigh*

>
> Anyway, your suggestion to speed up the search
>was a good one, except that it is even more crucial
>for him to fix the problem with quiessence-search
>first.

WTF is quiessence?

Do you mean quiescence?

Why would you search then anyway?

You don't search the same? Oh, you'll say it's different because you
don't know how long you have to search.

I still contend fix the strategy, then worry about the moving and
waiting to move stages. If you have a barge cobbled together with
garbage, it ain't going to be graceful, no matter how much chrome you
put on it.



> I am still beating up on his program in simple
>tactical skirmishes-- which indicates to me that it
>is not calculating them properly (whether looking
>ahead deeply or not). Speeding things up won't
>help here, any more than say, eating donuts more
>quickly will help unclog one's arteries.

I think we established that. A year ago. Hasn't changed.

>
>
> -- help bot


 
Date: 29 Jun 2008 21:42:08
From: Sanny
Subject: Re: 4 new Openings taught.
> > If you find any opening move missing let me know it.GetClubcan
> > remember first four moves. So it plays opening moves in 0 seconds
> > saving your time.
>
> =A0 About that: when there is only one legal
> move the program goes into a deep think,
> as usual. =A0But apart from figuring out that
> there is one and only one *legal* move,
> there's really nothing to think about, is
> there? =A0Just play the move on the board,
> and crank away on the opponent's time.

GetClub thinks when there is 1 forced move So that may be by thinking
higher depth it may give a Gambit of Rook/ Queen to win the game?

Or may be some move may be found which draws the game by repetitive
moves.

As many times some good move may look wrong at lower depth but only
when depth of search is increased we can see it's benefits.

I see in many games Zrbediah gives away its bishop/ Rook to win the
game So for that reasion all moves must be evaluated. Even if they are
giving loss at lower depth.

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html




 
Date: 29 Jun 2008 19:44:10
From: help bot
Subject: Re: 4 new Openings taught.

Sanny wrote:

> Today 4 new Openings were taught to GetClub Chess. So that it plays
> better against strong opponents.

> If you find any opening move missing let me know it. GetClub can
> remember first four moves. So it plays opening moves in 0 seconds
> saving your time.

About that: when there is only one legal
move the program goes into a deep think,
as usual. But apart from figuring out that
there is one and only one *legal* move,
there's really nothing to think about, is
there? Just play the move on the board,
and crank away on the opponent's time.

Not moving in this situation gives the
impression that the program is utterly
clueless about how to play chess.


-- help bot


  
Date: 03 Jul 2008 16:48:52
From: help bot
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 3, 4:48 am, Sanny <[email protected] > wrote:

> > I think you're talking about selective search.
> > I've read that those guys "lost" in the end,
> > that the thuggish brute-forcers somehow
> > came to dominate the field.
>
> Today again the game was improved. Now you will face tough Challenge
> at GetClub.
>
> Even Easy level will play remarkably better.

Thank goodness you are able to (just barely)
keep up with my own amazing improvement in
chess! Most others would have fallen behind
long ago, but not you. By implementing all
these incredible improvements, you now stand
alone, perhaps somewhere in the vicinity of
4,000,000-5,000,000 FIDE.

----------------------------------------------------------

Sanny, in order to give most of us humans
a decent challenge, all you need to do is get
the tactics part right. Instead of worrying over
pruning techniques to deepen the search,
you should think about researching what is
called "quiessence", or quietness-- the point
where you can safely stop looking further
ahead and return the position score back
down the line to an earlier node or whatever.

In the 1980s, Fidelity Electronics had an
entire series of portable chess computers,
with *very limited* speeds and memory
capacities which effectively got this part
down pat, and it showed in both results and
quality of play. They were terrible in the
endgame because of a dramatic speed-up
in simpler positions -- they messed up the
handling of thinking time -- but otherwise, I
believe that if someone had one of these
old-fangled contraptions in working order,
they could defeat the GetClub program in
a set match. No hash-tables, no Pentium
speed, just proper handling of tactics, and
a simplistic approach to piece-development
and control of the center. Even in the late
endgame, where both GC and the Chess
Challenger-x are at their worst, Fidelity had
you beat, for their contraption would force
the enemy King to the edge of the board as
part of its "control the center" obsession.

We're talking something like maybe 2 mHz
or less-- a modern PC can run rings around
these old farts, while simultaneously
watching MTV and eating peanut butter
flavored Captain Crunch. Heck-- even my
graphics chip could beat one of those old
things at chess.

I wish I had the expertise to help you, but
I don't. All I can do is point out the facts
and recommend that you look into certain
key areas (i.e. tactics).


-- help bot




   
Date: 04 Jul 2008 13:47:37
From: Patrick Volk
Subject: Re: When only 1 legal move, GetClub will not think more.
On Thu, 3 Jul 2008 16:48:52 -0700 (PDT), help bot
<[email protected] > wrote:

>On Jul 3, 4:48 am, Sanny <[email protected]> wrote:
>
>> > I think you're talking about selective search.
>> > I've read that those guys "lost" in the end,
>> > that the thuggish brute-forcers somehow
>> > came to dominate the field.
>>
>> Today again the game was improved. Now you will face tough Challenge
>> at GetClub.
>>
>> Even Easy level will play remarkably better.
>
> Thank goodness you are able to (just barely)
>keep up with my own amazing improvement in
>chess! Most others would have fallen behind
>long ago, but not you. By implementing all
>these incredible improvements, you now stand
>alone, perhaps somewhere in the vicinity of
>4,000,000-5,000,000 FIDE.
>
>----------------------------------------------------------
>
> Sanny, in order to give most of us humans
>a decent challenge, all you need to do is get
>the tactics part right. Instead of worrying over
>pruning techniques to deepen the search,
>you should think about researching what is
>called "quiessence", or quietness-- the point
>where you can safely stop looking further
>ahead and return the position score back
>down the line to an earlier node or whatever.
>
> In the 1980s, Fidelity Electronics had an
>entire series of portable chess computers,
>with *very limited* speeds and memory
>capacities which effectively got this part
>down pat, and it showed in both results and
>quality of play. They were terrible in the
>endgame because of a dramatic speed-up
>in simpler positions -- they messed up the
>handling of thinking time -- but otherwise, I
>believe that if someone had one of these
>old-fangled contraptions in working order,
>they could defeat the GetClub program in
>a set match. No hash-tables, no Pentium
>speed, just proper handling of tactics, and
>a simplistic approach to piece-development
>and control of the center. Even in the late
>endgame, where both GC and the Chess
>Challenger-x are at their worst, Fidelity had
>you beat, for their contraption would force
>the enemy King to the edge of the board as
>part of its "control the center" obsession.

Gee, but you forget the Atari 2600 chess program. It had 256 bytes of
RAM, and 2K. Advanced level it took 4 hours a move, but hey...


And you don't understand that to be tactically proficient you have to
understand the board. Score the board. Now, you can do that in an
O(2^n) fashion, or using some well-known memory efficient
constructions, cut that down to O(n log n).


>
> We're talking something like maybe 2 mHz
>or less-- a modern PC can run rings around
>these old farts, while simultaneously
>watching MTV and eating peanut butter
>flavored Captain Crunch. Heck-- even my
>graphics chip could beat one of those old
>things at chess.
>
> I wish I had the expertise to help you, but
>I don't. All I can do is point out the facts
>and recommend that you look into certain
>key areas (i.e. tactics).
>
>
> -- help bot
>


    
Date: 04 Jul 2008 13:13:04
From: Guest
Subject: Re: When only 1 legal move, GetClub will not think more.
"Patrick Volk" <[email protected] > wrote in message
news:[email protected]...
> On Thu, 3 Jul 2008 16:48:52 -0700 (PDT), help bot
> Gee, but you forget the Atari 2600 chess program. It had 256 bytes of
> RAM, and 2K. Advanced level it took 4 hours a move, but hey...

Actually, it had 128 bytes of RAM and some registers it could use for
storage.

As for the chess program, there were several versions with various ROM
sizes. The shipping versions fit into the 4k of rom space, but the original
versions didn't, which is why Atari developed the bank switching methods.

As a side note, I briefly emailed with the author a few years back and he
said he no longer has his original source or development stuff. He lost it
in a fire years ago.

He also told me that he originally wrote it in FORTRAN on a time share
system.

I think the source for at least one version of the Atari chess program was
disassembled and recommented into readable format. So if you are interested
you can browse it. (I never got around to downloading it and looking at
it.)


> And you don't understand that to be tactically proficient you have to
> understand the board. Score the board. Now, you can do that in an
> O(2^n) fashion, or using some well-known memory efficient
> constructions, cut that down to O(n log n).

Score the board in what way?

Just examining the board etc. won't result in a O(n^2) growth. Or O(N log
n) growth, for that matter.

The search isn't O(n^2) or O(n log n) either.

Nothing in chess programming is O(n^2) or O(n log n), actually? (Unless you
are counting the sort routine that programmers use to sort the move
ordering. Which is irrelevant.)

So what do you mean here?








----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
---= - Total Privacy via Encryption =---


     
Date: 04 Jul 2008 19:01:21
From: Patrick Volk
Subject: Re: When only 1 legal move, GetClub will not think more.
On Fri, 4 Jul 2008 13:13:04 -0500, "Guest" <[email protected] > wrote:

>"Patrick Volk" <[email protected]> wrote in message
>news:[email protected]...
>> On Thu, 3 Jul 2008 16:48:52 -0700 (PDT), help bot
>> Gee, but you forget the Atari 2600 chess program. It had 256 bytes of
>> RAM, and 2K. Advanced level it took 4 hours a move, but hey...
>
>Actually, it had 128 bytes of RAM and some registers it could use for
>storage.
>
>As for the chess program, there were several versions with various ROM
>sizes. The shipping versions fit into the 4k of rom space, but the original
>versions didn't, which is why Atari developed the bank switching methods.
>
>As a side note, I briefly emailed with the author a few years back and he
>said he no longer has his original source or development stuff. He lost it
>in a fire years ago.
>
>He also told me that he originally wrote it in FORTRAN on a time share
>system.
>
>I think the source for at least one version of the Atari chess program was
>disassembled and recommented into readable format. So if you are interested
>you can browse it. (I never got around to downloading it and looking at
>it.)

It's been a while since I mucked with the Atari. I did have the chess
gave however.

>
>
>> And you don't understand that to be tactically proficient you have to
>> understand the board. Score the board. Now, you can do that in an
>> O(2^n) fashion, or using some well-known memory efficient
>> constructions, cut that down to O(n log n).
>
>Score the board in what way?

Ok, I should have went into more detail. Scoring the position should
be constant. The process of determining moves is on the exponential
side.

>
>Just examining the board etc. won't result in a O(n^2) growth. Or O(N log
>n) growth, for that matter.
>
>The search isn't O(n^2) or O(n log n) either.
>
>Nothing in chess programming is O(n^2) or O(n log n), actually? (Unless you
>are counting the sort routine that programmers use to sort the move
>ordering. Which is irrelevant.)
>
>So what do you mean here?

If you're going to say 2^n is different than 3^n (or x^n), fine. Let's
just call it exponential and get on with life. Probably closer to
superexponential, and again not for scoring but the move list (I used
the term of search too loosely).

Think Sanny is being efficient? That is the issue. Think help bot is
right when he says Sanny might stumble on to a good way of making a
chess engine disregarding what is known about chess engines?




>
>
>
>
>
>
>
>
>----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
>http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
>---= - Total Privacy via Encryption =---


      
Date: 04 Jul 2008 18:49:53
From: Guest
Subject: Re: When only 1 legal move, GetClub will not think more.
"Patrick Volk" <[email protected] > wrote in message
news:[email protected]...
> On Fri, 4 Jul 2008 13:13:04 -0500, "Guest" <[email protected]> wrote:
>
>>> And you don't understand that to be tactically proficient you have to
>>> understand the board. Score the board. Now, you can do that in an
>>> O(2^n) fashion, or using some well-known memory efficient
>>> constructions, cut that down to O(n log n).
>>
>>Score the board in what way?
>
> Ok, I should have went into more detail. Scoring the position should
> be constant. The process of determining moves is on the exponential
> side.
>
>>
>>Just examining the board etc. won't result in a O(n^2) growth. Or O(N log
>>n) growth, for that matter.
>>
>>The search isn't O(n^2) or O(n log n) either.
>>
>>Nothing in chess programming is O(n^2) or O(n log n), actually? (Unless
>>you
>>are counting the sort routine that programmers use to sort the move
>>ordering. Which is irrelevant.)
>>
>>So what do you mean here?
>
> If you're going to say 2^n is different than 3^n (or x^n), fine. Let's

x^n is substantially different from n^2. That ^2 is fantastically weak
compared to what chess programs have to deal with.


> just call it exponential and get on with life. Probably closer to
> superexponential, and again not for scoring but the move list (I used
> the term of search too loosely).

More like O(width ^ Depth). (Although that is too simplistic, of course.)

Regular (& classic) programs would be along the lines of 6^Depth or so,
while modern state of the art would be around (2.x)^Depth.

The various ehnancements modern programs do can really make a difference, at
the risk of making the occasional error during the search.


Still, I'll grant you just a poor wording. I certainly do enough of that!


> Think Sanny is being efficient? That is the issue. Think help bot is
> right when he says Sanny might stumble on to a good way of making a
> chess engine disregarding what is known about chess engines?

I have not been following the GetClub stuff that much. Just off & on when I
come in here and lurk.

I don't really care that much about the program, although I do find it
somewhat intriquing to try and guess the program structure.

But no. I don't think Sanny is going in the right direction.

I could certainly be wrong. Especially since he hasn't discussed his
program's structure.

But if it's still making tactical errors, then there is something seriously
wrong.

Modern hardware (even that of just a few years ago, such as a 2gh P4) are
fast enough that a fairly simple program can get pretty darn deep and
wouldn't make tactical mistakes in at least the first 10 ply or so, even on
the faster levels.

If your tactical ability is poor or buggy, then you will never have a
successful chess program. (Probably. I'm leaving open the possibility of
some fantastic breakthrough in chess playing....(grin))

This is the same basic kind of problem that classic selective search
programs have had.

They may have significant positional knowledge, but their selective search
nature makes it easy for them to make tactical mistakes. And ultimately,
chess is a tactical game. Humans just are fast enough to be able to play it
tactically, so we do it positionally.


And ironically, a brute force program can actually have more knowledge than
a selective search program and be faster and search deeper. That was one of
the things Slate & Atkin discovered when they went from selective search in
Chess 3.6 to brute force in Chess 4.0 back in the early 70s. The cost of
picking the n-best moves is high in selective search but the benefits of
pruning are great in a full width program.

I'm a big fan of selective search, but I'm quite willing to admit the math
is not good for selective search. Brute force will see everything and wont
make tactical mistakes within its search depth. Selective search doesn't
see everything and it has a finite chance of completely ignoring the correct
move in the early part of the search and never being able to find the right
move.

And the state of the art pruning techniques used today can actually result
in a brute force having a narrower search width than a selective search
program would.



As to whether Sanny can stumble upon a good technique himself. Not enough
is known to judge what the program structure is really like.

It's possible he can manage enough to make a passable chess program. But it
would certainly be a heck of a lot easier if he learned more about chess
programs in general.

Even if he still wanted to go his own way later, at least he'd have the
background to be able to make an informed choice and know about potential
problems and so on.

I've kind of gotten the impression that GetClub chess is more selective
search or even knowledge based rather than a brute force. But I've never
played it and don't care to.






----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
---= - Total Privacy via Encryption =---


  
Date: 03 Jul 2008 01:48:41
From: Sanny
Subject: Re: When only 1 legal move, GetClub will not think more.
> =A0 I think you're talking about selective search.
> I've read that those guys "lost" in the end,
> that the thuggish brute-forcers somehow
> came to dominate the field.

Today again the game was improved. Now you will face tough Challenge
at GetClub.

Even Easy level will play remarkably better.

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html



  
Date: 02 Jul 2008 21:20:15
From: help bot
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 2, 11:29 pm, Patrick Volk <[email protected] > wrote:

> When you write a chess program, I think it's a pretty fair statement
> that you have to be knowledgeable in both fields, or at least know
> someone.

In reality, one can simply find a subroutine
which generates all legal moves from a
given position and "stick it in". That is what
I would do, so I could focus on other things.
You may not know this, but Sanny got
stuck there-- for a long, long time; in fact he
still hides the program's illegal moves.


> Hate to burst your bubble, but pruners and hash tables are hardly
> chess-centric programming techniques.

You must be a very young bot. In my day,
they had no hash-tables whatever, and it was
a big improvement when it finally came along.
The pruners have also improved dramatically
over time-- which is what I was referring to.


> The A-B table is straight out of
> game theory (which comes into more mundane things now such as search
> engines), and the hash table is pretty much used whenever you use any
> database (Apache uses them. Java has them

Nitwit! Those things did not exist when they
first started writing chess programs. Java?
Indians? You remind me of that punk kid who
writes about how inflation/stocks/home prices
are worse off than any time in the last couple
of years; hello! What's the big deal? Wake me
up when something /significant/ happens, like
we go back on the gold standard, or there's a
half-off cookie sale at Wal-mart!


> Why are they pretty standard in the chess game toolkit? Because
> chess has a massive amount of possible moves. Get 5-6 ply and your
> move set is at least in the billions. Every pieces adds at least an
> order of magnitude. A queen for example can have up to 28 different
> places it can move (a knight or king 8, a bishop/rook 14, and even a
> pawn has 4 possible moves).

Wrong. A pawn can have anywhere from
zero legal moves all the way up to twelve,
depending on the position of course.


> If you notice, every respectable chess program scores the position.
> Why? Because that score is used for A-B pruning. Basically, you don't
> trace down the less-than-favorible lines (basically, this determines
> the breadth. Depth is difficulty). Sure, you can put your queen next
> to your opponents' pawn to be taken in the next half ply, but if you
> have better moves, do you really want to analyze the permutations of
> that?

I think you're talking about selective search.
I've read that those guys "lost" in the end,
that the thuggish brute-forcers somehow
came to dominate the field.


> In other words, it's sorting. If you want to tell me a decent chess
> program can be made without sorting, I would have to say it's possible
> a roomful of monkeys can write Shakepeare. Possible, but EXTREMELY
> unlikely.

That's because no one has bothered to teach
them to write; you aren't even giving the poor
monkeys half a chance; it's obvious you are
unfamiliar with evolution. The fact is, even you
cannot write Shakespeare, so why are you
attacking the Monkeys? Giving them funny
looks? The fact remains they are too busy
singing, to put anybody down (unlike you).

Try this link: http://www.gorillatrade.com


> The fact that such things didn't occur to Sanny very much implies
> that his code is written as a journey. A journey that burns bridges.
> It's one thing to write a chess program, and not know one of the
> domains. It's either arrogant or ignorant to not know either.

I don't think you are aware that Sanny, so
he tells us, does not write the code himself.

He presumably pays some unknown
programmer or programmers to try to
implement his many "suggestions" for
improvements, and the result indicates an
unfamiliarity with chess-- even the basic
rules of the game. So, far from being a
poor programmer, Sanny is in reality not
a programmer at all.


> I'm not so sure Sanny would pass a Turing test at this point... If
> nothing else, he's optimistic. His website about management and stuff
> gave me a chuckle.
> Kind of reminds me of my 20's,

You're not still a teenager?


> where a bunch of us would sit around,
> and talk of starting up a company. But none of us really had an idea
> on what we wanted to do. That's where a plan comes in.

Sanny has sprung up at a time when Vishy
Anand has taken the world championship
title-- which may or may not relate to his
particular interest in chess. At a time when
Sam Sloan keeps screaming that the USCF
is "losing millions", it is funny to think that
/compared to them/, Sanny is likely millions
of rupees in the black, in spite of the quality
issues of his chess program.


-- help bot




   
Date: 09 Jul 2008 01:17:55
From: Sanny
Subject: Re: When only 1 legal move, GetClub will not think more.
> I've played it twice, and have better things to do than beta test.
> I'll play people on Pogo... better time controls.

Play again, As the game was improved today and it will play with
double strength now.

Play Chess at: http://www.GetClub.com/Chess.html

What is your rating? can you win the easy level now? I think Easy will
be very tough even for Help Bot to beat. But I find Help Bot always
get some way to win even after the game is improved.

It means Help Bot uses only partial energy to win and whenever
improvement is done he increases his strength to match the GetClub
level.

But this time it will be very hard to win for Help Bot.

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html





   
Date: 07 Jul 2008 21:03:52
From: help bot
Subject: Re: When only 1 legal move, GetClub will not think more.
On Jul 6, 11:29 pm, Patrick Volk <[email protected] > wrote:

> > You're splitting hairs; game theory/chess
> >game theory are the same thing, to me (i.e.
> >chess is a game).
>
> Game theory can be used in the routing of trains, or how to place
> your whiskey casks in a warehouse to move them the least (and store
> the most).

So then, whiskey is the reason? That explains
a lot. : >D




> Exactly why you need to be efficient. But if you're expecting a minute
> a move, and the program decides it needs to take 5, what do you expect
> the person to do?

The funny thing is, Sanny has just improved
the time management of his program just as
you say; it now allocates more time to the
tougher positions, and less to easier ones.
But this in no way aids in overcoming the
primary reason it loses to me or anyone
else (tactical errors).


> I've played it twice, and have better things to do than beta test.
> I'll play people on Pogo... better time controls.

Well then, that makes you quite the expert.


> When I work in a particular field, it's in my best interest to
> figure out what they do, and what the rules are. I don't mind reading
> to get the job done... Otherwise I'd be harping about how I took
> FORTRAN IV in college, like some people.

Wrong yet again. (I took it in high school.)
This pattern keeps repeating.



> Are you a chess programmer?

Can you read? I have written here over
and over and over that I wish I could help
Sanny with this, but I cannot... because___?

(Hint: it's not because I'm an arrogant
little snot like you.)


-- help bot





   
Date: 03 Jul 2008 12:54:16
From: Patrick Volk
Subject: Re: When only 1 legal move, GetClub will not think more.
On Wed, 2 Jul 2008 21:20:15 -0700 (PDT), help bot
<[email protected] > wrote:

>On Jul 2, 11:29 pm, Patrick Volk <[email protected]> wrote:
>
>> When you write a chess program, I think it's a pretty fair statement
>> that you have to be knowledgeable in both fields, or at least know
>> someone.
>
> In reality, one can simply find a subroutine
>which generates all legal moves from a
>given position and "stick it in". That is what
>I would do, so I could focus on other things.
> You may not know this, but Sanny got
>stuck there-- for a long, long time; in fact he
>still hides the program's illegal moves.

Subroutine is an interesting choice of term. So basic.

>
>
>> Hate to burst your bubble, but pruners and hash tables are hardly
>> chess-centric programming techniques.
>
> You must be a very young bot. In my day,
>they had no hash-tables whatever, and it was
>a big improvement when it finally came along.
>The pruners have also improved dramatically
>over time-- which is what I was referring to.

Hmm. I'm 41, been in the business since I've been 19, and know hash
tables were in chess programs in the 80's.

>
>
>> The A-B table is straight out of
>> game theory (which comes into more mundane things now such as search
>> engines), and the hash table is pretty much used whenever you use any
>> database (Apache uses them. Java has them
>
> Nitwit! Those things did not exist when they
>first started writing chess programs. Java?
>Indians? You remind me of that punk kid who
>writes about how inflation/stocks/home prices
>are worse off than any time in the last couple
>of years; hello! What's the big deal? Wake me
>up when something /significant/ happens, like
>we go back on the gold standard, or there's a
>half-off cookie sale at Wal-mart!

Sure, I'm a nitwit. Whatever you say, sparky. Hash tables were part
of the original Unix implementation.
Guess you like linear or exponential algorithms. Logarithmic
algorithms are too new-age... lol.

>
>
>> Why are they pretty standard in the chess game toolkit? Because
>> chess has a massive amount of possible moves. Get 5-6 ply and your
>> move set is at least in the billions. Every pieces adds at least an
>> order of magnitude. A queen for example can have up to 28 different
>> places it can move (a knight or king 8, a bishop/rook 14, and even a
>> pawn has 4 possible moves).
>
> Wrong. A pawn can have anywhere from
>zero legal moves all the way up to twelve,
>depending on the position of course.

Any piece can have a minimum of zero moves. Oh, and I forgot the
promotion.

Given that Sanny still has illegal moves, I'd say he has an issue with
this. Which was the point.

>
>
>> If you notice, every respectable chess program scores the position.
>> Why? Because that score is used for A-B pruning. Basically, you don't
>> trace down the less-than-favorible lines (basically, this determines
>> the breadth. Depth is difficulty). Sure, you can put your queen next
>> to your opponents' pawn to be taken in the next half ply, but if you
>> have better moves, do you really want to analyze the permutations of
>> that?
>
> I think you're talking about selective search.
>I've read that those guys "lost" in the end,
>that the thuggish brute-forcers somehow
>came to dominate the field.

A-B is selective search. Even the massively parallel solutions use
some form of selective search, if for nothing else, to explore the
best-looking lines first.
What's your point?

>
>
>> In other words, it's sorting. If you want to tell me a decent chess
>> program can be made without sorting, I would have to say it's possible
>> a roomful of monkeys can write Shakepeare. Possible, but EXTREMELY
>> unlikely.
>
> That's because no one has bothered to teach
>them to write; you aren't even giving the poor
>monkeys half a chance; it's obvious you are
>unfamiliar with evolution. The fact is, even you
>cannot write Shakespeare, so why are you
>attacking the Monkeys? Giving them funny
>looks? The fact remains they are too busy
>singing, to put anybody down (unlike you).

Sigh.... You never read Gamow. 1, 2, 3... Infinity is a pretty good
book, dare I say entertaining for programmers, and essential for
mathematicians.

>
>Try this link: http://www.gorillatrade.com

And the point is... what, why don't they just have a mutual fund?

>
>
>> The fact that such things didn't occur to Sanny very much implies
>> that his code is written as a journey. A journey that burns bridges.
>> It's one thing to write a chess program, and not know one of the
>> domains. It's either arrogant or ignorant to not know either.
>
> I don't think you are aware that Sanny, so
>he tells us, does not write the code himself.

He's familiar with it.

>
> He presumably pays some unknown
>programmer or programmers to try to
>implement his many "suggestions" for
>improvements, and the result indicates an
>unfamiliarity with chess-- even the basic
>rules of the game. So, far from being a
>poor programmer, Sanny is in reality not
>a programmer at all.

Whatever.

>
>
>> I'm not so sure Sanny would pass a Turing test at this point... If
>> nothing else, he's optimistic. His website about management and stuff
>> gave me a chuckle.
>> Kind of reminds me of my 20's,
>
> You're not still a teenager?
>
>
>> where a bunch of us would sit around,
>> and talk of starting up a company. But none of us really had an idea
>> on what we wanted to do. That's where a plan comes in.
>
> Sanny has sprung up at a time when Vishy
>Anand has taken the world championship
>title-- which may or may not relate to his
>particular interest in chess. At a time when
>Sam Sloan keeps screaming that the USCF
>is "losing millions", it is funny to think that
>/compared to them/, Sanny is likely millions
>of rupees in the black, in spite of the quality
>issues of his chess program.
>
>
> -- help bot
>


    
Date: 03 Jul 2008 22:00:04
From: Guest
Subject: Re: When only 1 legal move, GetClub will not think more.
"Patrick Volk" <[email protected] > wrote in message
news:[email protected]...
> On Wed, 2 Jul 2008 21:20:15 -0700 (PDT), help bot
> <[email protected]> wrote:
>
>>On Jul 2, 11:29 pm, Patrick Volk <[email protected]> wrote:
>>
>>> When you write a chess program, I think it's a pretty fair statement
>>> that you have to be knowledgeable in both fields, or at least know
>>> someone.
>>
>> In reality, one can simply find a subroutine
>>which generates all legal moves from a
>>given position and "stick it in". That is what
>>I would do, so I could focus on other things.
>> You may not know this, but Sanny got
>>stuck there-- for a long, long time; in fact he
>>still hides the program's illegal moves.
>
> Subroutine is an interesting choice of term. So basic.

It is a proper term. Not realistic considering data structures will be
different, and a little outdated, but still a proper term.




>>> Hate to burst your bubble, but pruners and hash tables are hardly
>>> chess-centric programming techniques.
>>
>> You must be a very young bot. In my day,
>>they had no hash-tables whatever, and it was
>>a big improvement when it finally came along.
>>The pruners have also improved dramatically
>>over time-- which is what I was referring to.
>
> Hmm. I'm 41, been in the business since I've been 19, and know hash
> tables were in chess programs in the 80's.

Hash tables were first used in Mac Hack VI back in 1966, when it was first
being written. That was the first truely significant chess program, as well
as the first one that used transposition tables.

Most significant mainframe chess programs also did hash tables.

Microcomputer programs weren't as eager to add them for some reason.

Even an 8 bit micro, with 64k could profit from them, but most didn't
bother.

Even many IBM PC programs, where you had 256k or 512k often didn't use them,
for some reason.

It's possible that many programers prefered to use refutation tables
instead. For shallow searches (less than 6 ply or so) a refuation table can
guide a search about as well as a trans table can, with a lot less memory.

Possibly they simply assumed that trans tables wouldn't be much of a benefit
for their limited search depths, considering the extra overhead involved.

Hyatt once published a paper and gave one example position where he had
CrayBlitz use just a few K bytes for the trans hash, pawn hash and king
hash. The hit rates were 19%, 90% and 80%.

Not great for the trans table itself, but pretty good for pawns & king
safety. But probably not too useful for micro programs because they often
had limited pawn structure & king safety terms because they had so little
time to perform the evaluations. A hash table would have let them put in
more knoledge, but they probably didn't realize that for some years.

I don't know what micro program was first to use trans tables, or when they
became common for micro programs.


This kind of reminds me of how Sargon 2 & 3 were. Sargon 2 still used a
static exchange evaluator. It wasn't until Sargon 3 that they switched to a
capture search and along with other improvements got a vastly faster search.
(Sargon 2 & 3 were only about a year apart, but they were sold commercially
for years. Kind of stupid to sell Sargon 2, but they did.)



>
>>
>>
>>> The A-B table is straight out of
>>> game theory (which comes into more mundane things now such as search
>>> engines), and the hash table is pretty much used whenever you use any
>>> database (Apache uses them. Java has them
>>
>> Nitwit! Those things did not exist when they
>>first started writing chess programs. Java?
>>Indians? You remind me of that punk kid who

AB alogorithm (AB table???) has a fuzzy origin.

Although we all know the official story of its origin, it's possible that
Aruther Samuel's checker program from the 50's used some similar idea.
However I certainly can not verify that.

Newel's 1955(ish) program used it. It provided such a massive performance
boost that was the reason that Herbert Simon made his famous 1957 statement
of "within ten years, a digital computer will be the world's chess champion,
unless rules bar it from competition."


>>> If you notice, every respectable chess program scores the position.
>>> Why? Because that score is used for A-B pruning. Basically, you don't
>>> trace down the less-than-favorible lines (basically, this determines
>>> the breadth. Depth is difficulty). Sure, you can put your queen next
>>> to your opponents' pawn to be taken in the next half ply, but if you
>>> have better moves, do you really want to analyze the permutations of
>>> that?
>>
>> I think you're talking about selective search.
>>I've read that those guys "lost" in the end,
>>that the thuggish brute-forcers somehow
>>came to dominate the field.
>
> A-B is selective search. Even the massively parallel solutions use
> some form of selective search, if for nothing else, to explore the
> best-looking lines first.
> What's your point?

Alpha-Beta is not selective. It's a pruning method that guarantees
identical results to a search without it. Just faster. (The same is true
of other search methods, such as Scout, NegaScout, PVS, etc. They all
guaranteed correct results.)

If you do a full width minimax search to 8 ply, you see everything. Nothing
is missed.

If you do a full width minimax search with Alpha-Beta to 8 ply, you see
everything. Nothing is missed. It's just faster, is all.

Things like null move, razoring, LMR and so are forms of selective search.
(As well as traditional selective search, such as MacHack VI, etc.)

Selective search does not guarnatee correct results. If you do a selective
search to 8 ply, it is possible for it to miss winning moves.

(Transposition tables can also cause some errors in the search, although
they aren't normally considered a form of 'selective' search but an
accelerator.)

My guess is that GetClub chess uses selective search. Probably the more
traditional definition, like MacHack VI, Awit, Chaos, etc.








----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
---= - Total Privacy via Encryption =---


 
Date: 29 Jun 2008 07:05:52
From: Sanny
Subject: 4 new Openings taught.
Today 4 new Openings were taught to GetClub Chess. So that it plays
better against strong opponents.

If you find any opening move missing let me know it. GetClub can
remember first four moves. So it plays opening moves in 0 seconds
saving your time.

Bye
Sanny

Play Chess at:http://www.GetClub.com/Chess.html



  
Date: 03 Jul 2008 01:51:20
From: Sanny
Subject: Re: When only 1 legal move, GetClub will not think more.
> Do you think you're attracting top-tier chess players with your
> program?

Not many, But now the Program plays very decently so even good players
will face the challenge.

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html



 
Date: 28 Jun 2008 04:00:29
From: Sanny
Subject: Re: Get Club beaten like carrots
> =A0 Nevertheless, it seems to get the tactics
> right up to a certain, albeit very shallow,
> depth. =A0Note how Zeb not only uses a much
> stronger engine, but he also takes White--
> as if to enjoy watching the carnage. =A0In the
> one game in which I let another engine
> duke it out with GC, I offered up QN odds
> as best I could (i.e. N-a3, N-b5, N-a7) and
> it was a reasonably interesting game for a
> while.
>

Game was improved today taking your advices. Thanks a lot for such
valuable information.

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html


 
Date: 28 Jun 2008 01:23:57
From: Sanny
Subject: Re: GetClub
> =A0 Sanny's estimates are always too close
> together; he likes the round number 100,
> and wishes his top levels to be equal to or
> higher than the braggers here in rgc. =A0But
> his program's results tell a very different
> story. =A0Because they don't like Sanny, the
> folks around here are highly motivated to
> win. =A0Even so, his program occasionally
> scores a victory or two, when the position
> is "just right" for his engine to not self-
> destruct. =A0It's greatest weakness is in the
> endgame; however...

Today again the game was improved a bit. So you will face a stronger
opponent now.

Play a game with Normal Level and tell me if you see the difference.

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html


 
Date: 27 Jun 2008 11:53:07
From:
Subject: Re: Off topic spam
> Only 3 players are able to win the higher Levels.
>
> 1. Help Bot - Beginner, Easy, Normal
> 2. Bonsai - Beginner, Easy, Normal
> 3. Zebediah Beginner, Easy, Normal, and Master and also Advance.


Don't you mean that only those 3 are willing to waste their time
playing
what ammounts to postal chess vs your very weak engine? Even your
very own site shows that your "Advanced" mode which takes hours per
move is +2 -48 and yet you assign it a "2400 rating on some imaginary
Sanny scale.

Although I know the effort will be fruitless I will offer two more
tidbits of
wisdom for you.

1) Virtually every program I have seen, including yours, has roughly
the
same strength at every time control. This presumes, of course, that
the
human is playing under the same time constraints. If anything, most of
them seem stronger to me at fast time controls.

2) The length of a games in moves is not even close to being an
accurate
gauge of the relative strength of the players. This tends to be more a
matter of style then strength.


 
Date: 27 Jun 2008 02:39:41
From: help bot
Subject: Re: Off topic spam
On Jun 27, 3:54 am, Sanny <[email protected] > wrote:

> > How can the so-called "advance" level, rumored to take as long as 6
> > hours for a move, play such a tactically terrible game as the one
> > above?

Actually, that game was a *strategical* crush;
no amount of number-crunching would have
saved Black from doom.


> Play with Master Level and see for yourself how it crushes you.
>
> Only 3 players are able to win the higher Levels.
>
> 1. Help Bot - Beginner, Easy, Normal

And Master, and Advance, though I may
die of old age.


> 2. Bonsai - Beginner, Easy, Normal

Bonsai's overall record is shown as 70--4;
maybe he just doesn't have the patience to
play the higher levels.


> 3. Zebediah Beginner, Easy, Normal, and Master and also Advance.

Zeb is obviously using another chess engine;
you're never going to be able to beat him.


> Rest other players cannot face even the Beginner Level..
>
> Beginner: 1800
> Easy: 1900
> Normal: 2000
> Master: 2100
> Advance: 2200

In the very next post in this thread, the
Advance level somehow miraculously jumps
to "2300+".

All these numbers are way too high. Look
at my most recent games-- a near-miniature
against the Normal level, demonstrating its
many inherent flaws:

http://www.getclub.com/playgame.php?id=DM21181&game=Chess

Another win where, this time, the program
does not merely self-destruct on its own:

http://www.getclub.com/playgame.php?id=DM21208&game=Chess

A tough fight against the *Advance* level,
in which the program managed to land a
decisive tactical blow for once, but then
botched the endgame terribly:

http://www.getclub.com/playgame.php?id=DM20686&game=Chess

In view of all this, I think the ratings of
1800--2300+ are way out of whack; just
chop off around 200 points or so from
those outlandish numbers:

Beginner: 1400
Easy: 1600
Normal: 1750
Master: 1900
Advance: 2000

Better still, get this thing going on FICS,
and let's stop with all the guessing games.


-- help bot







 
Date: 27 Jun 2008 00:55:28
From: Sanny
Subject: Re: Zebediah made the King dance from e8-a1.
> > But atlast has to surrender.
>
> Sanny, what's GM Atlast's rating?

I do not know may be 2300+. As Advance level plays as good as 2300+
rated player.

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html




 
Date: 27 Jun 2008 00:54:22
From: Sanny
Subject: Re: Off topic spam
> How can the so-called "advance" level, rumored to take as long as 6
> hours for a move, play such a tactically terrible game as the one
> above?

Play with Master Level and see for yourself how it crushes you.

Only 3 players are able to win the higher Levels.

1. Help Bot - Beginner, Easy, Normal
2. Bonsai - Beginner, Easy, Normal
3. Zebediah Beginner, Easy, Normal, and Master and also Advance.

Rest other players cannot face even the Beginner Level..

Beginner: 1800
Easy: 1900
Normal: 2000
Master: 2100
Advance: 2200

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html







 
Date: 27 Jun 2008 00:09:10
From: Wlodzimierz Holsztynski (Wlod)
Subject: Re: Zebediah made the King dance from e8-a1.
On Jun 26, 9:28 am, Sanny <[email protected] > wrote:

> But atlast has to surrender.

Sanny, what's GM Atlast's rating?

Wlod


 
Date: 26 Jun 2008 16:15:46
From: help bot
Subject: Get Club beaten like carrots
On Jun 26, 2:38 pm, "[email protected]" <[email protected] > wrote:

> How can the so-called "advance" level, rumored to take as long as 6
> hours for a move, play such a tactically terrible game as the one
> above?
>
> Based on this game can we even say that advance level is as good as
> 1800 ELO?

It is impossible to estimate accurately, since
we don't even know which engine Zeb is using.

The best way to get a reliable estimate is to
have the GetClub program play a number of
different opponents near its own level, some
games with White, some with Black, and in
several different openings.

A few obvious flaws which seem to repeat
themselves quite frequently:

1) ...B-a6 (as in this game);

2) Premature Queen moves (such as 1. d4 Nf6,
2. Qd3);

3) Preferring moves which lead to the King
getting trapped in the center.

The program is still buggy. I had a game
recently where it was not my turn to move
and the program declared itself the winner
in a losing position! My GC rating took a
nasty hit, I expect.

In another game, I miniaturized it (more
or less, depending on how you define it) in
a dull, closed position where it crammed
its pieces into a corner away from its own
King.

Nevertheless, it seems to get the tactics
right up to a certain, albeit very shallow,
depth. Note how Zeb not only uses a much
stronger engine, but he also takes White--
as if to enjoy watching the carnage. In the
one game in which I let another engine
duke it out with GC, I offered up QN odds
as best I could (i.e. N-a3, N-b5, N-a7) and
it was a reasonably interesting game for a
while.

My guess is that if you lined up a bunch
of human players and randomly stuck GC
somewhere in the middle, nobody could
tell which patzer was which, assuming
they were all facing a commercial chess
engine. I still see folks who play there
/anonymously/ and then, if and when they
get lucky, come here to brag. But where
are the real-life Class-x players now? now
that the program is no longer somewhere
around 1000 USCF? And lest we forget,
there is no real time limit for the human
opponents; how many times I would
have lost OTB if it weren't for this built-in
"safety"!

Somebody really ought to help Sanny out
with getting him fixed up on ICS or FICS;
not merely /telling him how/ to do it, but
actually git 'er done, so we all can see the
objective results.


-- help bot




 
Date: 26 Jun 2008 11:38:05
From: [email protected]
Subject: Re: Off topic spam
How can the so-called "advance" level, rumored to take as long as 6
hours for a move, play such a tactically terrible game as the one
above?

Based on this game can we even say that advance level is as good as
1800 ELO?


 
Date: 26 Jun 2008 10:21:47
From: Sanny
Subject: Re: Off topic spam
On Jun 26, 9:53=A0pm, "J.D. Walker" <[email protected] > wrote:
> Sanny wrote:
>
> > What do you think Advance Level was wrong that it lost its Rook?
>
> http://www.youtube.com/watch?v=3D-F6MwlEsdAA

So that Storm take away the Rook?

Bye
Sanny

Play Chess at: http://www.GetClub.com/Chess.html



 
Date: 26 Jun 2008 09:53:40
From: J.D. Walker
Subject: Re: Off topic spam
Sanny wrote:
>
> What do you think Advance Level was wrong that it lost its Rook?
>

http://www.youtube.com/watch?v=-F6MwlEsdAA
--

"Do that which is right..."

Rev. J.D. Walker