# This gets to ~19-20 reward in ~30 minutes / 4m steps on a m4.10xl instance
# TODO(rliaw): this has regressed in performance
pong-a3c:
    env: PongDeterministic-v4
    run: A3C
    config:
        # Works for both torch and tf.
        framework: tf
        num_workers: 16
        rollout_fragment_length: 20
        vf_loss_coeff: 0.5
        entropy_coeff: 0.01
        gamma: 0.99
        grad_clip: 40.0
        lambda: 1.0
        lr: 0.0001
        observation_filter: NoFilter
        preprocessor_pref: rllib
        model:
            use_lstm: true
            conv_activation: elu
            dim: 42
            grayscale: true
            zero_mean: false
            # Reduced channel depth and kernel size from default
            conv_filters: [
                [32, [3, 3], 2],
                [32, [3, 3], 2],
                [32, [3, 3], 2],
                [32, [3, 3], 2],
            ]