cql-halfcheetahbulletenv-v0:
    env: HalfCheetahBulletEnv-v0
    run: CQL
    pass_criteria:
        evaluation/episode_reward_mean: 400.0
        timesteps_total: 10000000
    stop:
        time_total_s: 3600
    config:
        # Use input produced by expert SAC algo.
        input: ["~/halfcheetah_expert_sac.zip"]
        actions_in_input_normalized: true

        soft_horizon: False
        horizon: 1000
        q_model_config:
            fcnet_activation: relu
            fcnet_hiddens: [256, 256, 256]
        policy_model_config:
            fcnet_activation: relu
            fcnet_hiddens: [256, 256, 256]
        tau: 0.005
        target_entropy: auto
        no_done_at_end: false
        n_step: 3
        rollout_fragment_length: 1
        replay_buffer_config:
          type: MultiAgentReplayBuffer
          learning_starts: 256
        train_batch_size: 256
        target_network_update_freq: 0
        min_train_timesteps_per_iteration: 1000
        optimization:
            actor_learning_rate: 0.0001
            critic_learning_rate: 0.0003
            entropy_learning_rate: 0.0001
        num_workers: 0
        num_gpus: 1
        metrics_smoothing_episodes: 5

        # CQL Configs
        min_q_weight: 5.0
        bc_iters: 20000
        temperature: 1.0
        num_actions: 10
        lagrangian: False

        # Switch on online evaluation.
        evaluation_interval: 3
        evaluation_parallel_to_training: true
        evaluation_num_workers: 1
        evaluation_duration: 10
        evaluation_duration_unit: episodes
        evaluation_config:
            input: sampler
        always_attach_evaluation_results: True