chess-tuning-tools
chess-tuning-tools copied to clipboard
A collection of scripts aimed at efficiently tuning chess engine parameters.
If cutechess-cli errors out or the executable was not found, [`parse_experiment_result`](https://github.com/kiudee/chess-tuning-tools/blob/ef33453b735abb67a68e6a7a3abb13d495bd341a/tune/local.py#L937) will produce cryptic errors. To improve the user experience, these errors messages should be made more informative.
For some parameter ranges (e.g., very large numbers) the plot tick labels can overlap. A potential solution could be to use rotated axis labels in these cases.
Currently, the distributed tuning framework has a lot of hardcoded values specific to Leela Chess Zero and Stockfish. Since the demand for distributed tuning has increased, this should be made...
Many options are difficult to understand, which is why a good set of notebooks could be useful to explain them: - [ ] Troubleshooting notebook: What problems can occur and...
Would be nice to be able to run position tests rather than games. Maybe even easier to accept output log from https://github.com/djChess2019/position-tester and create plots based on them?
* Chess Tuning Tools version: My fork = no opening book and normalize_y=True * Python version: 3.8.7 * Operating System: Linux ### Description I have noticed that the y-scales of...
> Is there any other reason for this than some computation time between iterations? That is, will the result be different compared to doing 25 times more iterations with 2...
The distributed tuning client uses its own implementation to run cutechess-cli matches. This should be replaced by the newer implementation here: https://github.com/kiudee/chess-tuning-tools/blob/435eafa9a31ab358463a1988eceb27c2486af441/tune/local.py#L243
The current landscape plot shows a good overview of the effects of the parameters, but it is not really possible to extract the uncertainty of the estimate of the global...