ludax-agents

Pretrained AlphaZero agents for games in the Ludax board game framework.

Installation

pip install ludax-agents

Or install together with Ludax and its GUI:

pip install 'ludax[gui,agents]'

What's included

Each bundled agent is a ResNet trained via self-play using Gumbel MuZero search. At inference time the agent runs 256 MCTS simulations per move.

Usage

The agents integrate automatically with the Ludax GUI — just install this package and an alphazero option will appear in the policy dropdowns for any supported game.

For programmatic use:

from ludax import LudaxEnvironment
from ludax.games import reversi
from ludax_agents import az_checkpoint_policy, get_checkpoint_path

env = LudaxEnvironment(game_str=reversi)
policy = az_checkpoint_policy(env, get_checkpoint_path("reversi"))

# policy(state_b, rng_key) -> action_b

Adding your own checkpoints

Checkpoints are produced by the AlphaZero trainer in examples/03-alpha-zero/. Copy the best iteration file and rename it <game_id>.ckpt (where game_id matches a key in ludax.games.__all__), then place it in src/ludax_agents/checkpoints/ and rebuild the package.

Links

• Ludax repository
• Ludax on PyPI
• Issues