Alphabeta is a compact chess engine with negamax alpha‑beta search (iterative deepening, TT, quiescence) and a swappable small ResNet value network trained via teacher‑assisted self‑play. Includes training scripts, match/eval tools, and a single‑file browser UI
Alphabeta — a small chess engine that learnsA from-scratch minimax chess engine with a walnut-and-ivory web UI — plus a neural network that learns to play by watching games and can take over as the engine’s evaluator. Train it overnight, then play against it in the browser or run it in a head-to-head match against the classic engine and watch its win rate climb.Two brains, one searchBoard representation, legal move generation, and check/mate detection use the python-chess library (correctly implementing chess rules — castling, en passant, threefold repetition — from scratch is its own project). Everything that makes this an engine rather than a rules library is hand-written in backend/engine.py:Minimax search with alpha-beta pruning (negamax formulation)Iterative deepening (search depth 1, 2, 3… within a time budget)Move ordering: MVV-LVA for captures, killer moves for quiet moves, and the transposition-table move firstTransposition table (Zobrist hashing via chess.polyglot.zobrist_hash)Quiescence search on captures to avoid the horizon effect
