Local tuner ~~~~~~~~~~~ - Add support for twosided resign adjudication. If set to true, both engines have to report an absolute score higher than the ``resign_score`` threshold (95). - Add support for ``engineX_restart`` (default "auto") which allows one to set the restart mode used by cutechess (95). - Add depth-based time control using ``engineX_depth`` (95). - Log the estimated draw rate of the current match (197).
0.9.2
------------------
Local tuner ~~~~~~~~~~~
- Add CLI flag ``--evaluate-points -p CSVFILE`` which allows the user to supply a CSV file with points the tuner should try before resuming normal operation. An optional integer column can be specified, which will be used to indicate the number of rounds to run each point for (91). - Log version of the tuner in the log output (191).
0.9.1
------------------
Local tuner ~~~~~~~~~~~
- Fix a bug where the Elo and optima plots were showing incorrect iterations, if the user reduced parameter ranges (183). - Only use scientific notation in the optima plots, if the parameter is optimized on a log scale (182).
0.9.0
------------------
Local tuner ~~~~~~~~~~~
- Add a plot which shows the optima predicted by the tuner across the iterations (172). This can be useful to gauge convergence. - Add a plot which shows the estimated Elo (+ confidence interval) of the predicted optima (176). - Tuner saves optima and their Elo performance (including standard deviation) to disk now (171).
0.8.3
------------------
- Fix compatibility with Python 3.7 (150).
0.8.2
------------------
Local tuner ~~~~~~~~~~~
* Add support for plotting one-dimensional optimization landscapes (34).