# This reaches ~20 reward in 50 minutes (6M train steps, 2M env steps) on a
# p3.2xlarge AWS instance.
# See https://app.wandb.ai/zplizzi/test/runs/ayuuhixr?workspace=user-zplizzi
# for training curves.
pong-apex:
    env: PongNoFrameskip-v4
    run: APEX
    config:
        # Works for both torch and tf.
        framework: tf
        target_network_update_freq: 20000
        num_workers: 4
        num_envs_per_worker: 8
        lr: .00005
        train_batch_size: 64
        gamma: 0.99