Sanity check on piece-value tables

[Henk]

Don't understand what a queen should do on h8 if black has castled long. Total insane. PSQ tables are not specific enough. And if you use many they are difficult to maintain.

[Dann Corbit]

"I was thinking of something a bit different: under the assumption that the underlying table should be symmetric..."

Heuristically, it does not appear to be true.
The shape of black's tables, collected from game data, is not identical to white's

[BeyondCritics]

How do you generate the heat for a certain field? There are several alternatives reasonable, i believe.
Looking at the values, i don't like them at all, they seem to be heavily distorted by irreversible effects within the game.
E.g. in most games black castles king side. For pawns on the king side it is therefore more likely, that there are left untouched, than for the pawns on the queen side. This seems to be reflected in your estimated values. But what happens, if your program decides to castle queen side?
Better you invent some methods to cast this "irreversible" effects out.

[lucasart]

Hi all,

I've been experimenting with generating material tables for use in Pigeon. Below is a heat map of some early results. Having never done this before, do numbers like these look sane?

(These tables are upside down (black's POV) because that's how my data was. Blue is more happy, red is more sad).



The data was generated using about 4m games (CCRL 40/40, CEGT blitz, and an export of millbase from SCID). From each game, I discarded non-quiet positions, and also threw away anything after a promotion (figuring that play would mostly be tactical at that point). The result was about 100m test positions. I then used logistic regression to bake the tables.

Some features make perfect sense, but something major I didn't expect was the pronounced kingside bias you see with the bishop and the queen. (?)

A possibility: I didn't filter the input games by player strength, so perhaps these are the ghosts of unbalanced games, where the strong player was able to storm the castle and mate before the weaker player's king could get free?

If you have any insight, please let me know!

You need to use different tables for opening and endgame, at least for the King. Indeed the King PST is the most valuable and sensitive (in elo terms). You want to incentivise the King to control the center in the endgame, but in the opening you should penalize a king in the center, and instead keep the King on or close to the optimal squares G1 or B1 (for white).

[lucasart]

Your approach is interesting though. What you could do, for King and Pawn PST is:
* restrict to pawn endgames only. Here you'll see King centralisation, and perhaps Pawn more advanced.
* ditto, but with the total piece material, or maybe one minor exchanged at most.

Then you have an opening, and an endgame PST, and you interpolate based on piece material.

[lucasart]

Also, a good way to start, is to focus only on King and Knight PST, and leave the others at zero. Those two are, by far, the most elo sensitive.

[Daniel Anulliero]

Just an idea à little off topic :
Would it be a good idea , decrease the psqt value (for pieces) if a square is defended / attacked by enemy pawns ?

[CheckersGuy]

Just an idea à little off topic :
Would it be a good idea , decrease the psqt value (for pieces) if a square is defended / attacked by enemy pawns ?

I dont know whether thats reasonably or not but I think you would end up with a heat map that represents more "safe" squares.

[BeyondCritics]

Have you read this?
https://chessprogramming.wikispaces.com/Square+Control

[brtzsnr]

Hi, Lucas!

Try to generate your psqt using quiet-labeled.epd from https://bitbucket.org/zurichess/tuner/downloads

These positions are many, very varied, quiet and labeled with the game result from stockfish vs stockfish. In my experience these are better than CCRL positions because the on CCRL there can been large difference in Elo between players.

If you do, please share you results.

Regards,