v5.0

ELO Change (self-play)

Elo | 113.74 +- 16.13 (95%)
SPRT | 60.0+0.60s Threads=1 Hash=64MB
LLR | 2.95 (-2.94, 2.94) [0.50, 2.50]
Games | N: 1474 W: 787 L: 321 D: 366
Penta | [27, 73, 235, 211, 191]

What's Changed

• Check extension in search by @Dannyj1 in #71
• Fix check extension causing less TT hits by @Dannyj1 in #72
• Late Move Reduction by @Dannyj1 in #73
• Bias tuner towards smaller evaluation values by lowering base values by @Dannyj1 in #74
• Tuner: Five-point Stencil Approximation, Train for multiple epochs, accuracy improvements & bug fixes by @Dannyj1 in #76
• Fix compilation by @Dannyj1 in #77

Full Changelog: v4.1...v5.0

v4.1

Elo Change (self-play)

Elo | 48.86 +- 6.75 (95%)
Conf | 60.0+0.60s Threads=1 Hash=64MB
Games | N: 7910 W: 3578 L: 2473 D: 1859
Penta | [296, 495, 1627, 882, 655]

What's Changed

• Fix SEE not handling captures properly by @Dannyj1 in #64
• QSearch and Move Generation Improvements (promotions in qsearch, evasions movegen) by @Dannyj1 in #65
• Fixed counter move history ordering by @Dannyj1 in #66
• Fix QSearch ignoring non-capture promotions by @Dannyj1 in #68
• Fixed some issues with insufficient material draws wrongfully being detected when there is sufficient material.

Full Changelog: v4.0...v4.1

v4.0

Another big update! This time I reworked the whole search to be less buggy and more accurate. There seem to be some weaknesses that still need work (especially the endgame), but the engine performs much better now.

ELO Change (self-play)

Elo | 293.33 +- 28.67 (95%)
SPRT | 60.0+0.60s Threads=1 Hash=64MB
LLR | 2.95 (-2.94, 2.94) [0.50, 2.50]
Games | N: 888 W: 689 L: 78 D: 121
Penta | [3, 7, 74, 96, 264]

What's Changed

• Evaluation of weak squares in mobility and defended minor pieces by @Dannyj1 in #60
• Search rework by @Dannyj1 in #61
• Time management and Null Move Pruning fixes by @Dannyj1 in #62

Full Changelog: v3.0...v4.0

v3.0

This is a pretty big update coming with a complete evaluation rewrite, improving performance, improvements to the gradient descent tuner and fixing a good amount of issues. Self-play shows a pretty good ELO change (although the real change against other engines is usually less).

ELO Change

Elo | 160.52 +- 30.39 (95%)
SPRT | 60.0+0.60s Threads=1 Hash=64MB
LLR | 1.33 (-2.94, 2.94) [0.50, 2.50]
Games | N: 542 W: 347 L: 113 D: 82
Penta | [9, 13, 88, 57, 104]

What's Changed

• Smarter/better time management by @Dannyj1 in #48
• Cleanup of search code by @Dannyj1 in #49
• Fix TT move sorting by @Dannyj1 in #50
• Simplify versioning scheme, no more "patch" versions by @Dannyj1 in #51
• Also add fail-low nodes to the TT by @Dannyj1 in #52
• Tuner improvements and fixes by @Dannyj1 in #55
• Evaluation rework by @Dannyj1 in #54
• Use central finite difference for gradient estimation by @Dannyj1 in #58

Full Changelog: v2.3.1...v3.0

v2.3.1

Elo Change

ELO | 44.87 +- 20.14 (95%)
CONF | 60.0+0.60s Threads=1 Hash=64MB
GAMES | N: 1020 W: 508 L: 377 D: 135

Notes

The engine is now a lot more conservative with it's time usage, maybe a bit too conservative. This will be tweaked/tuned in a future release.

What's Changed

• Fix outpost evaluation and pawn attacks not being taken into account during evaluation by @Dannyj1 in #43
• Properly handle the movestogo UCI parameter, fix some crashes when games last very long by @Dannyj1 in #45
• Time management improves & time forfeit fixes by @Dannyj1 in #47
• Multiple crash fixes
• Reduced the size of Transposition Table entries

Full Changelog: v2.3.0...v2.3.1

v2.3.0

What's Changed

• Fix TT entries not being saved correctly by @Dannyj1 in #37
• Bunch of performance improvements by @Dannyj1 in #39
• Fixed multiple evaluation bugs by @Dannyj1 in #40
• Start using aspiration windows from depth 3 instead of 2 by @Dannyj1 in #41

ELO | 32.39 +- 10.98 (95%)
SPRT | 60.0+0.60s Threads=1 Hash=64MB
LLR | 2.95 (-2.94, 2.94) [0.00, 3.00]
GAMES | N: 3464 W: 1702 L: 1380 D: 382

Full Changelog: v2.2.0...v2.3.0

v2.2.0

As I am finding a bit more time to work on this project, expect some more releases. This release contains some big performance improvements.

What's Changed

• Performance improvements by @Dannyj1 in #32
• Properly find optimal K by @Dannyj1 in #31
• Outpost evaluation by @Dannyj1 in #33
• Multiple bug fixes in move ordering by @Dannyj1 in #35
• Performance improvements by @Dannyj1 in #36

Full Changelog: v2.1.0...v2.2.0

v2.1.0

What's Changed

• Do an "in check" search extension before dropping into qSearch by @Dannyj1 in #24
• Null move pruning by @Dannyj1 in #25
• Dynamic LMR reduction by @Dannyj1 in #26
• Tuner tweaks by @Dannyj1 in #27
• Attacks near king evaluation rework by @Dannyj1 in #29
• Fixed some issues in the fail-soft framework

Full Changelog: v2.0.0...v2.1.0

Elo Change

ELO | 15.61 +- 11.65 (95%)
CONF | 60.0+0.60s Threads=1 Hash=64MB
GAMES | N: 2984 W: 1367 L: 1233 D: 384

v2.0.0

Small changelog but significant ELO gain. I now consider 2.0.0 stable enough to be a full release and it (almost) has feature parity with 1.0.

What's Changed

• Tune all evaluation parameters with a stochastic gradient descent tuner by @Dannyj1 in #22

Full Changelog: v2.0.0-alpha-3...v2.0.0

ELO | 71.54 +- 16.76 (95%)
SPRT | 10.0+0.10s Threads=1 Hash=64MB
LLR | 3.00 (-2.94, 2.94) [0.00, 3.00]
GAMES | N: 1640 W: 936 L: 603 D: 101

v2.0.0-alpha-3

What's Changed

• Better move ordering by @Dannyj1 in #11
• Late move reduction by @Dannyj1 in #12
• Improve king safety evaluation by @Dannyj1 in #14
• Pawn evaluation by @Dannyj1 in #16
• Performance improvements by @Dannyj1 in #17
• Fixed underflow in time management by @Dannyj1 in #18
• Fixed Late Move Reduction reducing early moves by @Dannyj1 in #20
• Check search extension by @Dannyj1 in #21

ELO | 309.00 +- 34.12 (95%)
CONF | 60.0+0.60s Threads=1 Hash=64MB
GAMES | N: 848 W: 708 L: 105 D: 35

Full Changelog: v2.0.0-alpha-2...v2.0.0-alpha-3