Minke - UCI chess engine
Author: Eduardo Marinho
Working in progress UCI compatible chess engine done in order to learn and to get fun. Based on NNUE evaluation, Bitboard board representation and a lot of search enhancements.
Minke 2.0.0 what's new?
Results against Minke 1.0.0
STC:
Results of minkeV2 vs minkeV1 (10+0.1, NULL, 16MB, 8moves_v3.pgn):
Elo: 80.86 +/- 17.37, nElo: 126.30 +/- 26.15
LOS: 100.00 %, DrawRatio: 41.59 %, PairsRatio: 4.08
Games: 678, Wins: 236, Losses: 81, Draws: 361, Points: 416.5 (61.43 %)
Ptnml(0-2): [6, 33, 141, 118, 41], WL/DD Ratio: 0.34
LLR: 2.95 (100.1%) (-2.94, 2.94) [0.00, 5.00]
SPRT ([0.00, 5.00]) completed - H1 was accepted
LTC:
Results of minkeV2 vs minkeV1 (60+0.6, NULL, 16MB, 8moves_v3.pgn):
Elo: 103.97 +/- 18.10, nElo: 176.44 +/- 28.93
LOS: 100.00 %, DrawRatio: 38.63 %, PairsRatio: 7.10
Games: 554, Wins: 195, Losses: 34, Draws: 325, Points: 357.5 (64.53 %)
Ptnml(0-2): [1, 20, 107, 115, 34], WL/DD Ratio: 0.13
LLR: 2.95 (100.1%) (-2.94, 2.94) [0.00, 5.00]
SPRT ([0.00, 5.00]) completed - H1 was accepted
Changes
Implementing Aspiration windows search #12
Improving Quiescence search #11
New neural net trained with roughly 13 billion Stockfish positions #10
Adding Improving heuristic #9
Implementing datagen #6
Implementing Internal Iterative Reductions #4
Quite a few bug fixes
Comments
Post a Comment