Arasan 22.1
Posted: Wed Jul 01, 2020 2:58 am
Arasan 22.1 is available from https://www.arasanchess.org.
Changes in Arasan 22.1:
1) Remove potential passer scoring.
2) Tighten validation of hash/killer moves
3) Fix Resign threshold option (CECP mode)
4) Simplify lmp threshold computation. No longer different for PV.
5) Modify formula for lmp count. Do not allow 0 as depth index.
6) Increase razor margin and reduce razoring depth.
7) Fixes to singular search code. Verify hash move before doing a singular search.
8) If singular extension is triggered, do not compute any additional extension, or reduction.
9) Implement bench command. Use bench command for PGO.
10) Update test tools to use pentanomial model for SPRT.
11) MultiPV bug fixes, including a fix for bad ponder move output in MultiPV mode.
12) Fix accept draw logic (CECP mode)
13) Do not update history on fail high of 1st quiet and low depth (idea from Ethereal).
14) Fix possible crash: when iterating through the search threads, ensure any that have
a null "work" pointer are skipped.
15) Allow ProbCut while in check, and restrict recursive Probcut
16) Optimization in seeSign() function
17) Misc. code cleanup
Some test results at 1:0+0.6 vs 22.0:
82 576 1409 803 130
6000 games, score: 52.69%
ELO: 18.72 +-5.3 (95%)
LOS: 100.0%
Changes in Arasan 22.1:
1) Remove potential passer scoring.
2) Tighten validation of hash/killer moves
3) Fix Resign threshold option (CECP mode)
4) Simplify lmp threshold computation. No longer different for PV.
5) Modify formula for lmp count. Do not allow 0 as depth index.
6) Increase razor margin and reduce razoring depth.
7) Fixes to singular search code. Verify hash move before doing a singular search.
8) If singular extension is triggered, do not compute any additional extension, or reduction.
9) Implement bench command. Use bench command for PGO.
10) Update test tools to use pentanomial model for SPRT.
11) MultiPV bug fixes, including a fix for bad ponder move output in MultiPV mode.
12) Fix accept draw logic (CECP mode)
13) Do not update history on fail high of 1st quiet and low depth (idea from Ethereal).
14) Fix possible crash: when iterating through the search threads, ensure any that have
a null "work" pointer are skipped.
15) Allow ProbCut while in check, and restrict recursive Probcut
16) Optimization in seeSign() function
17) Misc. code cleanup
Some test results at 1:0+0.6 vs 22.0:
82 576 1409 803 130
6000 games, score: 52.69%
ELO: 18.72 +-5.3 (95%)
LOS: 100.0%