personal experience regarding "self learning"

[deeds]

How to get statistics of experience data ?

There are many ways :

- when the engine loads its experience file


- with an external tool like eman_exp_stats

[deeds]

How to view/edit experience data ?

There are many ways :

- with the builtin "exp" command


- with the builtin "expex" command


- with an external tool like experience viewer

[deeds]

How to convert/export experience data ?

There are many ways :

- with an external tool like expToPGN

- with an external tool like expToBook

- with an external tool like plainToEXP

- with an external tool like expExtract

[deeds]

How to reinforce experience data ?

There are many ways :

- with an external tool like commenter_pgn

- with an external tool like ANALYSEUR_PGN

[deeds]

How do I make an opening book that matches an experience file ?

The main idea is that the opening book starts the game.
But it must contain the same openings learned by the engine.
So each time the engine will be out-of-book, it will play from positions already known by its experience file.
After enough games/experience data, i could append more "most effective" moves to the opening book.

learning.bin : book check

[deeds]

For tourneys at longer TC, there are 2 solutions :

- easiest way, to reinforce your existing experience data

Code: Select all

   # PLAYER                             :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppAvg  OppN
   1 40T 1m+1s Eman 8.30 nn-ad9b42354671    :       0   ----   404.5     800  50.6  110  589  101  73.6      -4     4
   5 40T 5m+5s Stockfish 15.0 x64 BMI2      :      -7     34    98.0     200  49.0   25  146   29  73.0       0     1

White advantage = -40.94 +/- 9.05
Draw rate (equal opponents) = 50.00 % +/- 0.00

semi-monumental achievement

- longest way, to train the engines with a TC 2m+2s
(at the moment, i don't know until what TC such learned experience data can resist on tourneys...)

It seems that "7 learning sessions at TC 1m+1s" + "2 learning sessions at TC 2m+2s" + "reinforced experience data at D40" can resist until TC 6m+6s.

[deeds]

How to analyze a game lost by a learning engine ?

You have to use this learning engine to analyze its own moves at greater depth and offer it other moves/lines.
Do not forget to analyze the positions containing the moves of its opponent in order to complete its PV.

Here is my routine :
experience viewer : analyse lost game

Then, we can relaunch a mini match based on this opening between the 2 same engines.
A few games will be enough to find a better defense.

[deeds]

The result only reflects the usage of the 'experience file' which is nothing else than a kind of book file... DS is out of 'book' (exp file) in move 11, saved time 62s already vs. SF.

3 misinformation at once, impressive !

During the game, some positions have no experience data, some moves come from the engine's evaluation, other moves come from the opponent's evaluation. So the result of the games is a combination of several factors.

As it was already explained here, an experience file is alive and updated in realtime. Furthermore, the opening books don't show the score and depth values of their moves so we don't know what book's moves were good or bad.

The experience files cannot be "out-of-book" because the engine adds new moves to them after each game. Even when the engine doesn't play an instant move from its experience file, the engine's evaluation is oriented by its experience data.

[deeds]

On your website, you seem to imply that engines with a learning feature know what the most probable pieces combination will be at played at the endgame depending on the chosen opening !? It's cheating isn't it ?

Yes I confirm that as soon as the opening phase is over, they know very quickly what the most probable pieces combination will be at the endgame phase.
Generally, they already expect the combination Rook + 2 pawns VS Rook + 2 pawns (KRPPvKRPP).
But if they have learned the chosen opening, the experience file will guide them more and more precisely on the most probable pieces combination.
See endgameStats on my website.

Regarding if it's cheating or not, I'm in a bad place, I don't participate in any tournament on the board or on chess servers so I don't know the official rules.
Probably the engines that choose their first moves "based on the most probable pieces combination at the endgame phase" have a slight advantage.
The use of an experience file also gives them a slight advantage even if they didn't learned the chosen opening.
The advantage is certain if they already learned the chosen opening.
The advantage becomes maximum if their experience data has been reinforced on the chosen opening.

[sarona]

On your website, you seem to imply that engines with a learning feature know what the most probable pieces combination will be at played at the endgame depending on the chosen opening !? It's cheating isn't it ?

Regarding if it's cheating or not, I'm in a bad place, I don't participate in any tournament on the board or on chess servers so I don't know the official rules.

How can the use of experience files/engines/books be considered cheating insofar as OTB play is concerned?

They are primarily used during pre-tournament or match preparation. There is nothing unethical about using engines before the event - as opposed to using them during an event (No, I am not accusing you, Hans ).