till75/rlcheckers

How till75/rlcheckers works

RLCheckers is a reproducible 5×5 checkers reinforcement-learning experiment. Users invoke the CLI to play a seeded random game, evaluate agents, train a Q-table, or run the three-seed gate. Those workflows depend on immutable board, move, and game-state rules; agents provide random, greedy, and Q-table play. Training supports fixed-random-opponent Q-learning and self-play against frozen snapshots, while evaluation runs seeded matchups, computes metrics, and writes artifacts. Edges reflect shown imports, documented behavior, or the CLI excerpt; several implementation details are unsampled, so uncertain wiring is omitted.

Main entry points: state.py Game state · gate.py Three-seed gate · cli.py CLI commands · arena.py Match arena

Components 16

Game rules

Board encoding src/rlcheckers/game/board.py
Legal moves src/rlcheckers/game/moves.py
Game state src/rlcheckers/game/state.py

Playing agents

Agent protocol src/rlcheckers/agents/protocol.py
Random agent src/rlcheckers/agents/random_agent.py
Greedy agent src/rlcheckers/agents/greedy_agent.py
Q-table agent src/rlcheckers/agents/q_agent.py

Training

Fixed-opponent Q-learning src/rlcheckers/training/train_q.py
Snapshot self-play src/rlcheckers/training/self_play.py

Evaluation

Match arena src/rlcheckers/evaluation/arena.py
Score metrics src/rlcheckers/evaluation/metrics.py
Evaluation artifacts src/rlcheckers/evaluation/artifacts.py
Three-seed gate src/rlcheckers/evaluation/gate.py

Command interface

CLI commands src/rlcheckers/cli.py
Config validation src/rlcheckers/config.py

Other components

Experimenter

Connections 28

Experimenter → CLI commands: runs commands README.md
CLI commands → Config validation: loads settings src/rlcheckers/cli.py
CLI commands → Match arena: runs matchups src/rlcheckers/cli.py
CLI commands → Fixed-opponent Q-learning: starts training src/rlcheckers/cli.py
CLI commands → Snapshot self-play: dispatches self-play src/rlcheckers/cli.py
CLI commands → Three-seed gate: runs gate src/rlcheckers/cli.py
Board encoding → Legal moves: supplies board src/rlcheckers/game/moves.py
Board encoding → Game state: wrapped by src/rlcheckers/game/state.py
Legal moves → Game state: applied by src/rlcheckers/game/state.py
Game state → Agent protocol: agent input src/rlcheckers/agents/protocol.py
Game state → Random agent: chooses moves from src/rlcheckers/agents/random_agent.py
Game state → Greedy agent: evaluated by src/rlcheckers/agents/greedy_agent.py
Game state → Q-table agent: valued by src/rlcheckers/agents/q_agent.py
Game state → Fixed-opponent Q-learning: training state src/rlcheckers/training/train_q.py
Game state → Snapshot self-play: training state src/rlcheckers/training/self_play.py
Fixed-opponent Q-learning → Q-table agent: updates Q-table src/rlcheckers/training/train_q.py
Snapshot self-play → Q-table agent: freezes snapshots src/rlcheckers/training/self_play.py
Match arena → Agent protocol: requests moves src/rlcheckers/evaluation/arena.py
Match arena → Game state: plays games src/rlcheckers/evaluation/arena.py
Score metrics → Match arena: summarizes results src/rlcheckers/evaluation/metrics.py
Evaluation artifacts → Match arena: writes matchup data src/rlcheckers/evaluation/artifacts.py
Evaluation artifacts → Score metrics: writes summaries src/rlcheckers/evaluation/artifacts.py
Three-seed gate → Snapshot self-play: trains seeds src/rlcheckers/evaluation/gate.py
Three-seed gate → Match arena: evaluates matchups src/rlcheckers/evaluation/gate.py
Three-seed gate → Score metrics: aggregates scores src/rlcheckers/evaluation/gate.py
Three-seed gate → Greedy agent: tests baseline src/rlcheckers/evaluation/gate.py
Three-seed gate → Random agent: tests baseline src/rlcheckers/evaluation/gate.py
Three-seed gate → Q-table agent: evaluates learned agent src/rlcheckers/evaluation/gate.py

Drawn by GitDiagram on November 11, 2026 from the repository's file tree, README and source files. Read as Markdown ·