The author wrote: " As before, the engine supports SMP, Polyglot opening books, and Gaviota endgame tablebases. The new release includes a better optimization framework, an improved evaluation function, and tuned parameters. I only have a Mac at the moment, so I could not test the engine's playing strength using cutechess-cli, but I expect it to be around 2400-2500 Elo when using 4 search threads on a quad-core machine based on test suite results. The zip file contains the jar file, a batch script and a shell script for running the engine in UCI mode (depending on whether you are testing on Windows or Unix), a Polyglot opening book, the Gaviota tablebases for up to 4 men, and an external configuration file that allows you to play with the engine's parameters. The only thing you really need is the jar file; the book, the endgame tablebases, and the parameters file are all optional."
New version of K2 chess engine is out, the ELO gain is about 80 points.
In this release almost all eval terms have been tuned with Texel's method using positions from Zurichess testset. Also, some speedup and search improvements (LMR, NMR, futility pruning) was made.
Thanks!
Those 1.23 and 1.27 versions are Ronald Chastain Windows builds ( 1.23 and 1.27 ) in that list.
I did some stability check. Sapeli usually scores 80% - 85% in ( 40 moves/1 second ) games against Fairy. There seems not to be bad bugs in my dynamic hash implementation. But you never know. Also Sapeli used to eat 1.5 GB RAM. Now the default is 1 GB. Which should be reasonable.
Finished game 499 (Sapeli 1.29 vs Fairy-Max 5.0b): 1-0 {White mates}
Score of Sapeli 1.29 vs Fairy-Max 5.0b: 404 - 53 - 42 [0.852] 499
Started game 500 of 1000 (Fairy-Max 5.0b vs Sapeli 1.29)
Finished game 500 (Fairy-Max 5.0b vs Sapeli 1.29): 0-1 {Black mates}
Score of Sapeli 1.29 vs Fairy-Max 5.0b: 405 - 53 - 42 [0.852] 500
Started game 501 of 1000 (Sapeli 1.29 vs Fairy-Max 5.0b)
^CTerminating process of engine Sapeli 1.29(0)
Finished game 501 (Sapeli 1.29 vs Fairy-Max 5.0b): * {No result}
Score of Sapeli 1.29 vs Fairy-Max 5.0b: 405 - 53 - 42 [0.852] 500
Elo difference: 304.07 +/- 39.51
Finished match
New version of K2 chess engine is out, the ELO gain is about 80 points.
In this release almost all eval terms have been tuned with Texel's method using positions from Zurichess testset. Also, some speedup and search improvements (LMR, NMR, futility pruning) was made.
After Norbert pointed me in the right direction, I managed to build cutechess and did some preliminary testing of DETROID 1.0.1 on my i9-8950HK using a 40/20s time control which I assume translates to longer time controls on an AMD X2 4600+. The results of the gauntlet tournament I ran against Mediocre 0.5, CuckooChess 1.12, and Karballo 1.8 suggest an Elo rating of 2500+ for DETROID 1.0.1 running on 4 cores. I would be curious to see how this holds up under proper testing conditions.
After Norbert pointed me in the right direction, I managed to build cutechess and did some preliminary testing of DETROID 1.0.1 on my i9-8950HK using a 40/20s time control which I assume translates to longer time controls on an AMD X2 4600+. The results of the gauntlet tournament I ran against Mediocre 0.5, CuckooChess 1.12, and Karballo 1.8 suggest an Elo rating of 2500+ for DETROID 1.0.1 running on 4 cores. I would be curious to see how this holds up under proper testing conditions.
Seems about right. I am in the middle of a gauntlet and I expect it to be close to 2450 on 1 core.
