Experiment results

In this section, we'll take a look at the results of our multi-step training process

The baseline agent

To train the agent, run Chapter17/01_a2c.py with the optional --cuda flag to enable GPU and required -n option with the experiment name used in TensorBoard and in a directory name to save models.

Chapter17$ ./01_a2c.py --cuda -n tt
AtariA2C (
  (conv): Sequential (
    (0): Conv2d(2, 32, kernel_size=(8, 8), stride=(4, 4))
    (1): ReLU ()
    (2): Conv2d(32, 64, kernel_size=(4, 4), stride=(2, 2))
    (3): ReLU ()
    (4): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1))
    (5): ReLU ()
  )
  (fc): Sequential (
    (0): Linear (3136 -> 512)
    (1): ReLU ()
  )
  (policy): Linear (512 -> 4)
  (value): Linear (512 -> 1)
)
4: done 13 episodes, mean_reward=0.00, best_reward=0.00, speed=99.96
9: done 11 episodes, mean_reward=0.00, best_reward=0.00, speed=133.25
10: done 1 episodes, mean_reward=1.00, best_reward=1.00, speed=136.62
13: done 9 episodes, mean_reward...