The optimisation from lantonov

lantonov / optimisation Goto Github PK

View Code? Open in Web Editor NEW

Implementation of ROCK* algorithm (Gaussian kernel regression + natural gradient descent) for optimisation

License: GNU General Public License v3.0

Perl 27.81% Python 57.59% Jupyter Notebook 14.61%

optimisation's People

Contributors

Stargazers

Watchers

optimisation's Issues

cutechess -> pentanomial

The scripts bayes_opt and cutechess_batch make an incorrect assumption in the calculation of the pentanomial frequencies. Namely that the game pairs are consecutive in a cutechess match. That's not the case with concurrency>1. E.g.:

Finished game 1 (test vs base): 0-1 {Black wins by adjudication}
Finished game 2 (base vs test): 1/2-1/2 {Draw by 3-fold repetition}
Finished game 3 (test vs base): 1-0 {White wins by adjudication}
Finished game 5 (test vs base): 1-0 {White wins by adjudication}
Finished game 4 (base vs test): 1-0 {Black loses on time}
Finished game 8 (base vs test): 1-0 {White wins by adjudication}
Finished game 9 (test vs base): 1-0 {White wins by adjudication}
Finished game 6 (base vs test): 0-1 {White loses on time}
Finished game 10 (base vs test): 1-0 {White wins by adjudication}
Finished game 7 (test vs base): 0-1 {White loses on time}
Finished game 11 (test vs base): 1-0 {White wins by adjudication}
Finished game 12 (base vs test): 0-1 {Black wins by adjudication}
Finished game 13 (test vs base): 0-1 {Black wins by adjudication}
Finished game 15 (test vs base): 0-1 {Black wins by adjudication}
Finished game 16 (base vs test): 1-0 {White wins by adjudication}
Finished game 17 (test vs base): 1-0 {White wins by adjudication}
Finished game 14 (base vs test): 0-1 {White loses on time}
Finished game 18 (base vs test): 1/2-1/2 {Draw by 3-fold repetition}
Finished game 19 (test vs base): 1-0 {White wins by adjudication}
Finished game 20 (base vs test): 1-0 {White wins by adjudication}

Recommend Projects