*estimated

Features

Supports FRC, also known as Chess960

Search

The search is a standard alpha-beta search with the following enhancements:

• Iterative deepening
• Quiescence search
• Aspiration windows
• Principal variation search
• Transposition table
  • Move ordering
  • Cutoffs
  • Static evaluation correction
• MVV and SEE ordering of captures
• History Heuristic (standard history, 1 ply conthist, and noisy history)
• Reverse futility pruning
• Null move pruning
• Razoring
• Mate distance pruning
• History pruning
• Singular extensions
  • Double extension
  • Multicut
  • Negative extensions
• Correction history

Evaluation

The evaluation is done using a neural net trained entirely on self play games from zero knowledge using the excellent open source bullet neural network trainer. The architecture of the network is (768x8hm -> 1280)x2 -> 1x8

Build instructions

1. Get the network with git submodule update --init --depth 1
2. Install zig (0.14.0)
3. zig build --release=fast --prefix <installation path> (for example --prefix ~/.local will put pawnocchio in ~/.local/bin/pawnocchio) The Makefile is only intended to be used for testing on Openbench.

Licensing

• The code is licensed under the GPLv3 license. Full text can be found in LICENSE in the project root
• The assets are licensed under the CC-BY-ND 4.0 license. Full text can be found in assets/LICENSE

Credit

• Pyrrhic by Andrew Grant for tablebase probing, under the MIT license.
• Jackal by snekkers for inspiration styling this readme