Interesting adaptive engine

[noctiferus]

Citation from a paper by the authors of Bayeschess:
"BayesChess: A computer chess program based on Bayesian networks
Antonio Fernández, a, and Antonio Salmeróna,
Abstract
In this paper, we introduce a chess program able to adapt its game strategy to its opponent, as well as to adapt the evaluation function that guides the search process according to its playing experience. The adaptive and learning abilities have been implemented through Bayesian networks. We show how the program learns through an experiment consisting on a series of games that point out that the results improve after the learning stage.
Keywords: Bayesian networks; Adaptive learning; Computer chess

The engine, compiled by JA, is here:
http://www.mediafire.com/?kaaefdzypz0i8os

[LetoAtreides82]

It is a UCI engine or is it stand-alone?

[Uly]

He said "compiled by [Jim Ablett]", so it must be UCI or Winboard.

One always dreams with a truly learning engine like this, as in theory, after enough games against Rybka or Stockfish it would become very strong. Up to date, I've seen 3 kinds of learning:

Eval learning (as in Shredder): The engine continues where it left off the analysis last time. Future analysis is propagated to the root to avoid mistakes on the same positions. Extremely useful and powerful for analysis, but only works on specific positions where it was used, once a new positions arises, all the learning has no effect.

Result learning (as in ProDeo): The engine changes the evaluation of the moves depending on game results. Would make the engine give very accurate scores on opening positions (e.g. only engine I've seen from scratch to like c5 as best answer against e4). As good as a good book.

Monkey See Monkey Do (as in RomiChess): The engine doesn't change evaluation at all, it just plays instantly moves played by the opponent based on game results (discovering the moves that win and playing them, avoiding moves that lose). Quick 200 elo gain after enough games (probably due to time advantage). Can be very hurtful against weak engines. 0 use for analysis.

I hope I have enough time to check Bayesian learning.

[Hagen]

Until I actually see this posted on JA's site I'm not downloading this.

[Uly]

Good point, apparently Jim doesn't acknowledges this engine on his site.

Downloading never hurts, but I won't touch the zip file until I get confirmation from someone else that the file is clean and works.

[noctiferus]

This is the post where I found the info.

"http://www.mediafire.com/?kaaefdzypz0i8os This is a link for Bayeschess, compiled by JA, an adaptive chess program. It adapts to the playing style of it's opponent and opponent strength. It is in spanish, with it's own gui and it does play well, no apparent bugs. It seems to learn very quickly. Attached here is a pdf file with a full program description. It was to long to copy and paste, 12 pages. I have been playing against it and it does adapt and change it's play. The d / l is 8 to 10meg. The source files and pdf file are included with the d / l. The program loads slow, several minutes, but plays very quickly. This is the real deal, what sharpie chess engine was suppose to be. Fascinating to a nerd like me. fossil"

Note: it comes with his own GUI. i doubt it's a UCI or WB eng...
I posted it as soon as I saw, not yet tested.

However:
1- I tested the file with VirusTotal: 0 warning out of 42 AntiVirus.
2- The methodological paper is published on a well recognized international journal
3- it appears in Mediafire as being in "Jim Ablett's shared files".
I'll contact JA for having an official confirmation.

Comment: I don't think it's a top engine, IMHO it's just for fun, in order to have a competitor at our level, whichever it can be...

[noctiferus]

This is the mail I sent to JA:

Dear Jim,
first of all, thanks for your great job.

Now, to the point: I posted a link to Bayeschess that is loaded in Mediafire under "JA shared files", while, however, it does not appear in your webpage.

Some people raised doubts about the authenticity of this compile.
Would you be so kind to confirm, or disconfirm, that it is one of your compiles?

Thanks so much in advance for your attention.
Enrico

PS: May I post your reply on OpenChess forum?

[noctiferus]

Jim's answer to the mail above:

Hi Enrico,

Yes it is & Yes you may.

Jim.

[Hagen]

I'm wondering if this is the engine that was alleged to be able to do a new kind of heuristic analysis for playing chess and allows it to "self improve". There was a thread I think it was CC on the other site that talked about this engine and was dismissed as a joke. If this engine is the real deal...this could prove very interesting for the computer chess community. I'm still not downloading this until JA puts it on his site.

[DaveD]

I'm still not downloading this until JA puts it on his site.

Jim posted a link to this program at the CCC Forum ...

http://www.talkchess.com/forum/viewtopi ... bayeschess