Function test_agent

rl3/ddpg.py:191–205

(num_episodes=5)

  test_returns = []
  def test_agent(num_episodes=5):
    t0 = datetime.now()
    n_steps = 0
    for j in range(num_episodes):
      s, episode_return, episode_length, d = test_env.reset(), 0, 0, False
      while not (d or (episode_length == max_episode_length)):
        # Take deterministic actions at test time (noise_scale=0)
        test_env.render()
        s, r, d, _ = test_env.step(get_action(s, 0))
        episode_return += r
        episode_length += 1
        n_steps += 1
      print('test return:', episode_return, 'episode_length:', episode_length)
      test_returns.append(episode_return)
    # print("test steps per sec:", n_steps / (datetime.now() - t0).total_seconds())


  # Main loop: play episode and train