cartpole-ars:
    env: CartPole-v1
    run: ARS
    stop:
        sampler_results/episode_reward_mean: 150
        timesteps_total: 1000000
    config:
        # Works for both torch and tf.
        framework: torch
        noise_stdev: 0.02
        num_rollouts: 50
        rollouts_used: 25
        num_workers: 2
        sgd_stepsize: 0.01
        noise_size: 25000000
        eval_prob: 0.5