dqn_agent.json

{
    "type": "dqn_agent",

    "update_mode": {
        "unit": "timesteps",
        "batch_size": 64,
        "frequency": 4
    },
    "memory": {
        "type": "replay",
        "capacity": 10000,
        "include_next_states": true
    },

    "optimizer": {
        "type": "clipped_step",
        "clipping_value": 0.1,
        "optimizer": {
            "type": "adam",
            "learning_rate": 1e-3
        }
    },

    "discount": 0.99,
    "entropy_regularization": null,
    "double_q_model": true,

    "target_sync_frequency": 1000,
    "target_update_weight": 1.0,

    "actions_exploration": {
        "type": "epsilon_anneal",
        "initial_epsilon": 0.5,
        "final_epsilon": 0.0,
        "timesteps": 10000
    },

    "saver": {
        "directory": null,
        "seconds": 600
    },
    "summarizer": {
        "directory": null,
        "labels": [],
        "seconds": 120
    },
    "execution": {
        "type": "single",
        "session_config": null,
        "distributed_spec": null
    }
}