# @OldAPIStack
# To generate training data, first run:
# $ ./train.py --run=PPO --env=CartPole-v1 \
#      --stop='{"timesteps_total": 50000}' \
#      --config='{"output": "dataset", "output_config": {"format": "json", "path": "/tmp/out", "max_num_samples_per_file": 1}, "batch_mode": "complete_episodes"}'
cartpole-bc:
    env: CartPole-v1
    run: BC
    stop:
        timesteps_total: 500000
    config:
        # Works for both torch and tf.
        framework: torch
        enable_rl_module_and_learner: false
        enable_env_runner_and_connector_v2: false
        # In order to evaluate on an actual environment, use these following
        # settings:
        evaluation_num_env_runners: 1
        evaluation_interval: 1
        evaluation_config:
            input: sampler
        # The historic (offline) data file from the PPO run (at the top).
        input: dataset
        input_config:
            format: json
            paths: /tmp/out