v0.4.0

v0.4.0

This release further adds and refines search features, while increasing the net size to a hidden layer of 512.

Estimated Elo

~3510

Selfplay progression test:

Elo   | 138.38 +- 14.91 (95%)
Conf  | 10.0+0.10s Threads=1 Hash=8MB
Games | N: 1004 W: 434 L: 54 D: 516
Penta | [1, 24, 170, 208, 99]

New Features

• History LMR
• More LMR in non-PV nodes
• Continuation history (CMH + FMH etc..)
• Capture history heuristic
• Multicut (using singular search result)
• Expositor time management
• Killer bonus
• Cutnode LMR
• Increase to bucketed 512 hidden layer
• Butterfly history
• Precomputed LMR table
• SPSA 1.4k iterations
• Double extensions
• SPSA 2.3k iterations
• Aspiration reduction

v0.3.0

v0.3.0

This release fixes important bugs while adding and refining search features, creating another large improvement in strength.

Estimated Elo

~3400

Selfplay progression test:

Elo   | 223.29 +- 15.19 (95%)
Conf  | 10.0+0.10s Threads=1 Hash=8MB
Games | N: 1274 W: 743 L: 21 D: 510
Penta | [1, 9, 109, 303, 215]

New Features

• Check extensions
• TT prefetch
• Efficiently update NN
• NMP static evaluation difference
• Singular extensions
• Using TT in QS
• Soft time management fix
• LMP break moveloop

v0.2.0

v0.2.0

This release brings many new search features, creating a large improvement in strength.

Estimated Elo

~3200

Selfplay progression test:

Elo   | 203.41 +- 22.68 (95%)
Conf  | 10.0+0.10s Threads=1 Hash=8MB
Games | N: 1014 W: 686 L: 152 D: 176
Penta | [17, 24, 108, 124, 234]

New Features

• Keep TT bestmove
• Use partial search results
• LMR log formula
• Killer moves
• Aspiration windows
• PVS SEE pruning (quiets)
• Internal Iterative Reduction (IIR)
• QS SEE pruning
• SEE ordering
• Improving heuristic
• LMP
• Adaptive NMP
• LMR improving
• LMR for bad captures

v0.1.0

v0.1.0

This is the initial release of Exocet, a successor to Peacekeeper, and serves as a bug-free and OpenBench compatible base to build off of.

Estimated Elo

2900-2950

New Features

• Alphabeta
• Iterative Deepening
• Quiescence Search
• MVV/LVA capture ordering
• TT move ordering
• PVS
• NMP
• RFP
• TT cutoffs
• LMR
• History heuristic