I uploaded my training data at https://bitbucket.org/zurichess/tuner/d ... s/tuner.7z . This is very similar, but larger than the training data I used to train zurich graubuenden. Retraining with the new data resulted in ~8Elo gain:
8900 @ 40/15+0.15
2979 - 2766 - 3155
If you find it helpful, please let me know in a private message not to spam this forum. The contents of the README.txt included in the archive:
Code: Select all
What is in this archive ==== violent.epd - 1500k EPDs of quiet and violent positions. quiet.epd - 730k EPDs of quiet positions, subset of violent.epd. quiet-labeled.epd - 725k EPDs of quiet positions, subset of quiet.epd, annotated with a game result. How to use the data ==== quiet-labeled.epd can be used with the Texel Tuning Method https://chessprogramming.wikispaces.com/Texel%27s+Tuning+Method. Same training data is used to train zurichess http://www.zurichess.xyz. How was the data generated ==== 1. 75000 games were played between three slightly different versions of zurichess using 2moves_v1.pgn opening book to ensure high play variability. 2. From each game 20 positions were sampled from the millions of positions evaluated by the engine during the game play. This resulted in 1500k random positions which were stored in violent.epd. 3. From the set were removed all positions on which quiescence search found a wining capture. The remaining positions were stored in quiet.epd. 4. Next, from each quiet position a game using Stockfish 080916 was played. The results were stored in quiet-labeled.epd.