Lozza 2.0

"Lozza 2.0 is pretty much unchanged except that material, PSTs and all evaluation params were tuned in one go using Peter O's 'Texel' method" — the classic +1/-1 tuner, which ran for weeks. About +150 Elo over 1.18 at 60+1. Node.js is now required for offline use (jsUCI support removed). Tuner: texeltune.js.

Changes

• Add imbalance terms when no pawns.
• Tune all eval params.
• Swap mate and draw testing order in search.
• Do LMR earlier.
• Add draft bench command.
• Use pre generated random numbers using https://github.com/davidbau/seedrandom.
• Use depth^3 (>=beta), depth^2 (>=alpha) and -depth (< alpha) for history.
• Add -ve history scores for moves < alpha.
• Don't do LMP in a pvNode. We need a move!
• Remove support for jsUCI.
• Tune piece values and PSTs.
• Rearrange eval params so they can be tuned.
• Simplify phase and eval calc.
• Don't try and reduce when in check (optimisation).

Rating

CCRL 40/40: 2607 · CCRL Blitz: 2554

Usage

Use lozza.js with Node.js in any UCI chess GUI (Arena, Cute Chess, etc.), or in a web page/web worker.

Binaries

While Lozza is intended for web use, I can supply binaries on request. Please email me op12no2@gmail.com noting the target platform; thanks.