-
Notifications
You must be signed in to change notification settings - Fork 2.3k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
base WDL model on material count and normalize evals dynamically #5121
base WDL model on material count and normalize evals dynamically #5121
Conversation
The output from the fitting script is
|
The lower limit of a material count of And here the distribution of the WDL raw data: Looking at this plot, I guess we could use |
based on these graphs, I would say 10 is a good choice. Extending too much to small material count impacts the quality of the fit for the more relevant material counts. |
This PR proposes to change the parameter dependence of Stockfish's internal WDL model from full move counter to material count. In addition it ensures that an evaluation of 100 centipawns always corresponds to a 50% win probability at fishtest LTC, whereas for master this holds only at move number 32. See also official-stockfish#4920 and the discussion therein. The new model was fitted based on about 340M positions extracted from 5.6M fishtest LTC games from the last three weeks, involving SF versions from e67cc97 (SF 16.1) to current master. The involved commands are for [WDL_model](https://github.com/official-stockfish/WDL_model) are: ``` ./updateWDL.sh --firstrev e67cc97 python scoreWDL.py updateWDL.json --plot save --pgnName update_material.png --momType "material" --momTarget 58 --materialMin 10 --modelFitting optimizeProbability ``` The anchor `58` for the material count value was chosen to be as close as possible to the observed average material count of fishtest LTC games at move 32 (`43`), while not changing the value of `NormalizeToPawnValue` compared to the move-based WDL model by more than 1. The patch only affects the displayed cp and wdl values. closes official-stockfish#5121 No functional change
This PR proposes to change the parameter dependence of Stockfish's internal WDL model from full move counter to material count. In addition it ensures that an evaluation of 100 centipawns always corresponds to a 50% win probability at fishtest LTC, whereas for master this holds only at move number 32. See also #4920 and the discussion therein.
The new model was fitted based on about 340M positions extracted from 5.6M fishtest LTC games from the last three weeks, involving SF versions from e67cc97 (SF 16.1) to current master.
The involved commands are for WDL_model are:
The anchor
58
for the material count value was chosen to be as close as possible to the observed average material count of fishtest LTC games at move 32 (43
), while not changing the value ofNormalizeToPawnValue
compared to the move-based WDL model by more than 1.The patch only affects the displayed cp and wdl values.
No functional change.