trpo_sb.json

{
    "type": "trpo_agent",
    "batch_size": 20,
    "learning_rate": 1e-3,

    "discount": 0.99,
    "entropy_regularization": null,
    "gae_lambda": 0.97,
    "likelihood_ratio_clipping": null,

    "baseline_mode": "states",
    "baseline": {
      "type": "mlp",
      "sizes": [64, 64, 64]
    },
    "baseline_optimizer": {
        "type": "multi_step",
        "optimizer": {
            "type": "adam",
            "learning_rate": 1e-3
        },
        "num_steps": 5
    }
}