New Houdini

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

User avatar
Houdini
Posts: 1471
Joined: Tue Mar 16, 2010 12:00 am

Re: re: win adjudication/Re: New Houdini

Post by Houdini »

Lyudmil Tsvetkov wrote:Is not this 'simple 3-fold draw score' a kind of contempt?
It's just about avoiding silly 3-fold repetition when the position is equal.
Why do you call this "contempt", what definition do you use for "contempt"?
Lyudmil Tsvetkov
Posts: 6052
Joined: Tue Jun 12, 2012 12:41 pm

Re: re: win adjudication/Re: New Houdini

Post by Lyudmil Tsvetkov »

Houdini wrote:
Lyudmil Tsvetkov wrote:Is not this 'simple 3-fold draw score' a kind of contempt?
It's just about avoiding silly 3-fold repetition when the position is equal.
Why do you call this "contempt", what definition do you use for "contempt"?
Robert, please be so kind to answer the 2 other specific questions I addressed to you, and then I will tell you what my definition of contempt is.

OK, my definition for 'contempt' would be "any kind of a software code in a chess program that deliberately makes the program weaker in terms of picking up optimal moves for the purpose of prolonging the game and thus creating an opportunity for the program to achieve better results against opponents of lower and considerably lower strength"

But people are not very much concerned about definitions, what they would like to know is what would H score with drawscore=0 and drawscore=-7 in the current rapid, where it is leading SF, while in stage 3 it trailed behind.
User avatar
Houdini
Posts: 1471
Joined: Tue Mar 16, 2010 12:00 am

Re: re: win adjudication/Re: New Houdini

Post by Houdini »

Lyudmil Tsvetkov wrote:But people are not very much concerned about definitions, what they would like to know is what would H score with drawscore=0 and drawscore=-7 in the current rapid, where it is leading SF, while in stage 3 it trailed behind.
It shouldn't make any difference.
Note that Houdini is exactly the same engine with identical settings in the Rapid, there is no difference with Stage 3. Perceived performance variations are just a roll of the dice.
mjlef
Posts: 1494
Joined: Thu Mar 30, 2006 2:08 pm

Re: re: win adjudication/Re: New Houdini

Post by mjlef »

Houdini wrote:
Lyudmil Tsvetkov wrote:and what do the -5cps(stage 3) and -7cps(rapid) scores at drawing nodes/3-fold repetitions mean then?
That's a simple 3-fold draw score to avoid silly move repetitions. The score depends on the material that is still on the board. The 3-fold repetition score only operates in a narrow band around 0.00 and doesn't change whether queens or other pieces are kept on the board.

The real contempt used by Houdini and Komodo has the effect of modifying the piece values ("my pieces are worth slightly more than your pieces") so that exchanges are avoided unless they are really favorable. This contempt works all the time, not just in a very narrow band around 0.00.
Actually, Komodo's attempt does not modify the value of its own pieces. It does something else to discourage early piece captures and avoid draws (with a positive contempt), or exchange pieces and seek draws (with a negative contempt). But I can see how people might think it modifies piece scores.
User avatar
Rebel
Posts: 7027
Joined: Thu Aug 18, 2011 12:04 pm
Full name: Ed Schröder

Re: re: win adjudication/Re: New Houdini

Post by Rebel »

mjlef wrote:
Houdini wrote:
Lyudmil Tsvetkov wrote:and what do the -5cps(stage 3) and -7cps(rapid) scores at drawing nodes/3-fold repetitions mean then?
That's a simple 3-fold draw score to avoid silly move repetitions. The score depends on the material that is still on the board. The 3-fold repetition score only operates in a narrow band around 0.00 and doesn't change whether queens or other pieces are kept on the board.

The real contempt used by Houdini and Komodo has the effect of modifying the piece values ("my pieces are worth slightly more than your pieces") so that exchanges are avoided unless they are really favorable. This contempt works all the time, not just in a very narrow band around 0.00.
Actually, Komodo's attempt does not modify the value of its own pieces. It does something else to discourage early piece captures and avoid draws (with a positive contempt), or exchange pieces and seek draws (with a negative contempt). But I can see how people might think it modifies piece scores.
It's a nice idea anyway.
Lyudmil Tsvetkov
Posts: 6052
Joined: Tue Jun 12, 2012 12:41 pm

Re: re: win adjudication/Re: New Houdini

Post by Lyudmil Tsvetkov »

Houdini wrote:
Lyudmil Tsvetkov wrote:But people are not very much concerned about definitions, what they would like to know is what would H score with drawscore=0 and drawscore=-7 in the current rapid, where it is leading SF, while in stage 3 it trailed behind.
It shouldn't make any difference.
Note that Houdini is exactly the same engine with identical settings in the Rapid, there is no difference with Stage 3. Perceived performance variations are just a roll of the dice.
Now this is simply not true.

I respect very much your work on Houdini, but you simply state too many things the way they are not.

Nothing changed between H_stage3 & H_rapid?
Certainly, the -5 score was changed to -7 automatically in some way, by H itself, by the TCEC directors, by the TCEC software?

Komodo and Houdini will stick together here, of course, as commercial engines having implemented contempt, but both drawscore and asymmetric material scores are some form of contempt.

I actually do not see any difference at all between a code like:

Code: Select all

value[my_queen]=value[opponent_queen] +20;
value[my_rook]=value[opponent_rook] +10;
value[my_pawn]=value[opponent_pawn] +1; etc., for asymmetric material piece value contempt
and

Code: Select all

score+=7; for drawscore
Both are forms of contempt and do apply with all 4 of the following conditions of my contempt definition (no one actually disagreed with it):

1. they do avoid choosing the best possible move in a range of situations
2. they do prolong the game, creating further possibilities for the stronger engine
3. they do artificially inflate the rating performance against a bigger pool of lower-rated or considerably lower-rated engines
4. they do artificially deflate the rating performance against engines of equal strength or stronger engines

Can anyone deny any of the above statements?

Now, if you are using any form of contempt or drawscore, you would like to use it solely for the purpose of inflating engine score versus a larger pool of weaker opponents; you would not like your engine picking worse moves at times, neither you would like your engine losing against engines of equal strength.

So, the sole purpose of any form of contempt, no matter how it is implemented, is to artificially inflate rating performance in a larger pool of lower-rated opponents.

I wonder why no one mentions that even a very narrow band of drawscore(say, just 7cps), allows the engine to prune away in its search a large number of lines that would otherwise be returned as draws, thus prolonging the game artificially.

I am not certain what is the exact performance difference between say, material contempt=30, and drawscore=7. There certainly might be a difference, but I am not sure how you quantify it; in any case, both artificially inflate your rating, while, absurdly, weakening your play.

I bet SF with drawscore=7 would have performed somewhat better in the rapid, avoiding a number of draws, do not you think?
User avatar
Houdini
Posts: 1471
Joined: Tue Mar 16, 2010 12:00 am

Re: re: win adjudication/Re: New Houdini

Post by Houdini »

mjlef wrote:Actually, Komodo's attempt does not modify the value of its own pieces. It does something else to discourage early piece captures and avoid draws (with a positive contempt), or exchange pieces and seek draws (with a negative contempt). But I can see how people might think it modifies piece scores.
Nor does Houdini.
There is a difference between the simple description of the principle I gave ("my pieces are better than yours") and the actual implementation in Houdini which is - of course - not done by modifying the piece values, but by correcting the score based on the game phase.

I have noticed your tendency to describe Komodo's working in an obscure way ("it does something different"), so as to make it appear more complex than it actually is...
Last edited by Houdini on Thu Oct 13, 2016 11:31 am, edited 1 time in total.
User avatar
Houdini
Posts: 1471
Joined: Tue Mar 16, 2010 12:00 am

Re: re: win adjudication/Re: New Houdini

Post by Houdini »

Lyudmil Tsvetkov wrote:Now this is simply not true.

I respect very much your work on Houdini, but you simply state too many things the way they are not.

Nothing changed between H_stage3 & H_rapid?
Certainly, the -5 score was changed to -7 automatically in some way, by H itself, by the TCEC directors, by the TCEC software?
The 3-fold value is not -5 nor -7, it varies. See my post above: "the score depends on the material that is still on the board.". You could easily have concluded this from observing Houdini's games in TCEC.
The Houdini that plays in the Rapid is rigorously the same as in Stage 3.
Lyudmil Tsvetkov
Posts: 6052
Joined: Tue Jun 12, 2012 12:41 pm

Re: re: win adjudication/Re: New Houdini

Post by Lyudmil Tsvetkov »

Houdini wrote:
Lyudmil Tsvetkov wrote:Now this is simply not true.

I respect very much your work on Houdini, but you simply state too many things the way they are not.

Nothing changed between H_stage3 & H_rapid?
Certainly, the -5 score was changed to -7 automatically in some way, by H itself, by the TCEC directors, by the TCEC software?
The 3-fold value is not -5 nor -7, it varies. See my post above: "the score depends on the material that is still on the board.". You could easily have concluded this from observing Houdini's games in TCEC.
The Houdini that plays in the Rapid is rigorously the same as in Stage 3.
Ok, even if this were true, it is still contempt.
shrapnel
Posts: 1339
Joined: Fri Nov 02, 2012 9:43 am
Location: New Delhi, India

Re: re: win adjudication/Re: New Houdini

Post by shrapnel »

Houdini wrote:I have noticed your tendency to describe Komodo's working in an obscure way ("it does something different"), so as to make it appear more complex than it actually is...
Heh heh...let me get the Popcorn.
This slugfest should be interesting ! :D
Watch out Lefler...looks like ol' Houdart knows all about your Komodo ! So no more trying to be too mysterious ! :D
i7 5960X @ 4.1 Ghz, 64 GB G.Skill RipJaws RAM, Twin Asus ROG Strix OC 11 GB Geforce 2080 Tis