autoresearch-poker

Autonomous research loop for Texas Hold'em poker AI. An agent modifies train.py, trains a DQN for 5 minutes, evaluates against a random opponent, keeps/discards based on bb/hand, and loops forever.

Quick Start

uv sync
uv run train.py

Files

• prepare.py — Fixed: environment setup, evaluation harness, constants (DO NOT MODIFY)
• train.py — Agent modifies this: DQN config, network, training loop
• program.md — Agent instructions