v3.0.0 - Search improvements and bug fixes

This new release of byte-knight focuses on search improvements and bug fixes. Overall, there is a ~200 elo gain over v2.0.0.

Search

• Implemented late move reductions
• Implemented internal iterative reductions
• Implemented null move pruning

Other changes and fixes

• byte-knight now outputs the correct principle variation during search instead of just a best move
• Added transposition table probing and storing in qsearch.
• Updated search bench FEN suite
• Small code refactoring to make code more ergonomic
• More mate in X tests
• Fixed bug when negative times are given with go wtime/btime commend
• Fixed bug with history bonus/malus calculation in certain situations
• Added way to track node types during search
• Increased base search bench depth to 8

Comparison to v2.0.0

STC

Elo   | 197.30 +- 7.95 (95%)
Conf  | 8.0+0.08s Threads=1 Hash=16MB
Games | N: 5000 W: 2855 L: 286 D: 1859
Penta | [6, 104, 519, 1057, 814]
https://pyronomy.pythonanywhere.com/test/2755/

LTC

Elo   | 200.24 +- 10.84 (95%)
Conf  | 40.0+0.40s Threads=1 Hash=128MB
Games | N: 2500 W: 1421 L: 121 D: 958
Penta | [0, 37, 272, 545, 396]
https://pyronomy.pythonanywhere.com/test/2756/

---

The following release notes are auto generated:

What's Changed

• chore: v2.0.0 release back-merge by @DeveloperPaul123 in #75
• feat: add release target for apple silicon by @DeveloperPaul123 in #79
• feat: "properly" track nodes during search by @DeveloperPaul123 in #81
• chore: change default search-bench depth to 8 by @DeveloperPaul123 in #83
• feat: implement iir by @DeveloperPaul123 in #86
• chore: update bench FEN suite by @DeveloperPaul123 in #87
• feat: NMP implementation and board fixes by @DeveloperPaul123 in #88
• feat: tt probe in qs by @DeveloperPaul123 in #89
• feat: store moves to tt in QS by @DeveloperPaul123 in #90
• feat: add late move pruning (LMR) by @DeveloperPaul123 in #95
• chore: code cleanup and ergonomics improvements by @DeveloperPaul123 in #96
• chore: add small LMR offset by @DeveloperPaul123 in #97
• chore: reduce LMR scaling factor by @DeveloperPaul123 in #98
• chore: add more mate in x tests by @DeveloperPaul123 in #99
• fix: clamp negative times with go uci command by @DeveloperPaul123 in #101
• fix: output correct PV by @DeveloperPaul123 in #103
• chore: move history calc to a function by @DeveloperPaul123 in #104

Full Changelog: v2.0.0...v3.0.0

v2.0.0 - Search and evaluation improvements

The next release of byte-knight! This release sees improvements in both search and evaluation as well as some bug squashing. Overall ~250/265 elo gain for STC/LTC respectively over v1.0.0.

Search

• Implemented principle variation search
• History heuristic with history gravity
• Aspiration windows
• Reverse futility pruning

Evaluation

• Re-scaled and fixed minor issues related to MVV-LVA move ordering scoring
• Re-wrote most of the evaluation implementation
• Added an HCE tuner
• Tuned PSQT values for the engine using the lichess-big-3 data set

Other changes and fixes

• Fixed a bug related to king captures
• Moved to Rust 2024 (stable)
• Improved CI and added ability to generate code coverage metrics (integrated with CodeCov)
• Updated dev dependencies (mostly suggested by Dependi)

Comparison to 1.0.0

STC

Elo   | 251.95 +- 12.77 (95%)
Conf  | 8.0+0.08s Threads=1 Hash=16MB
Games | N: 3006 W: 2102 L: 238 D: 666
Penta | [6, 52, 245, 472, 728]
https://pyronomy.pythonanywhere.com/test/1831/

LTC

Elo   | 264.99 +- 18.94 (95%)
Conf  | 40.0+0.40s Threads=1 Hash=128MB
Games | N: 1500 W: 1070 L: 106 D: 324
Penta | [5, 24, 100, 244, 377]
https://pyronomy.pythonanywhere.com/test/1832/

---

These release notes are auto-generated by Github.

What's Changed

• chore: minor tweaks by @DeveloperPaul123 in #38
• feature: fast modulo for tt indexing by @DeveloperPaul123 in #39
• feat: principal variation search by @DeveloperPaul123 in #45
• fix: compare best to correct alpha by @DeveloperPaul123 in #46
• chore: simplify mvv-lva calculation by @DeveloperPaul123 in #47
• chore: rescale mvv-lva (again) by @DeveloperPaul123 in #48
• fix: mvv-lva scaling and minor issue with scoring for king captures by @DeveloperPaul123 in #49
• feat: add history heuristic by @DeveloperPaul123 in #50
• feat: use history gravity formula for history bonus by @DeveloperPaul123 in #51
• feat: implement aspiration windows by @DeveloperPaul123 in #53
• chore: re-write evaluation in preparation for tuning by @DeveloperPaul123 in #58
• chore: update deps and move to rust 2024 by @DeveloperPaul123 in #59
• fix: properly scaling eval and perspective by @DeveloperPaul123 in #65
• feat: tune on lichess big 3, from 0, 100k epochs by @DeveloperPaul123 in #66
• feat: add reverse futility pruning by @DeveloperPaul123 in #67
• feat: rework just file and add code coverage generation by @DeveloperPaul123 in #73

Full Changelog: v1.0.0...v2.0.0

v1.0.0 - Initial Release!

This is the first release of byte-knight. Based on play against stash, the estimated ELO of byte-knight is around ~1800. The following features are implemented in this version:

Search

• Iterative deepening
• Negamax with a/b pruning
• Quiescent search
• Move ordering using basic MVV/LVA table
• PeSTO based evaluation with tapering using the PeSTO PSQT values
• Transposition table (used for move ordering and cutoffs)

Game

I've written my own board representation complete with legal and pseudo-legal move generation as well as magic bitboards for sliding piece attacks. Performance is acceptable and will likely be improved on in the future.

---

These release notes are auto-generated by Github.

What's Changed

• Fix generating sliding attacks (include edges) by @DeveloperPaul123 in #1
• Fix perft and move generation by @DeveloperPaul123 in #2
• feat: legal move generation by @DeveloperPaul123 in #3
• feat: add open bench support by @DeveloperPaul123 in #4
• fix: issues in move ordering and search by @DeveloperPaul123 in #5
• fix: standing pat eval was backwards by @DeveloperPaul123 in #6
• feat: re-enable qsearch by @DeveloperPaul123 in #7
• fix: issue with pin ray calculation in certain situations by @DeveloperPaul123 in #10
• chore: rename byte_board -> chess by @DeveloperPaul123 in #11
• feat: implement pesto evaluation using psqt tables by @DeveloperPaul123 in #13
• feat: add support for basic search cancelation by @DeveloperPaul123 in #14
• chore: update how tt-move scoring is done. by @DeveloperPaul123 in #16
• fix: open-bench crashes by @DeveloperPaul123 in #17
• fix: clippy suggestions by @DeveloperPaul123 in #15
• refactor: simplify the input hander by @DeveloperPaul123 in #18
• fix: remove unsafe code by @DeveloperPaul123 in #20
• chore: minor tweaks and fixes by @DeveloperPaul123 in #21
• refactor: restructure project by @DeveloperPaul123 in #23
• chore: update ci runner images to be explicit by @DeveloperPaul123 in #25
• chore: add documentation by @DeveloperPaul123 in #27
• fix: store tt table per game, not per search by @DeveloperPaul123 in #28
• chore: nuke search by @DeveloperPaul123 in #29
• feat: re-add qsearch to main search by @DeveloperPaul123 in #30
• feat: implement (again) transposition table by @DeveloperPaul123 in #31
• feat: shrink the tt entry object size by @DeveloperPaul123 in #32
• fix: bug in psqt by @DeveloperPaul123 in #34
• chore: add unit tests for MVV/LVA eval for ordering by @DeveloperPaul123 in #35
• chore: prep for 1.0.0 release by @DeveloperPaul123 in #36

New Contributors

• @DeveloperPaul123 made their first contribution in #1

Full Changelog: https://github.com/DeveloperPaul123/byte-knight/commits/v1.0.0