Reinforcement Learning on Pong Atari 2600 (test)
20Average Total RewardDQN
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DQNpolicy=epsilon-greedy, epsilon=0.052013.12 | 20 | — | |
| Humanagent=expert human player2013.12 | -3 | — | |
| Contingencypolicy=epsilon-greedy, epsilon=0.052013.12 | -17 | — | |
| Sarsapolicy=epsilon-greedy, epsilon=0.052013.12 | -19 | — | |
| Randompolicy=random2013.12 | -20.4 | — | |
| DQN Bestpolicy=epsilon-greedy, epsilon=0.05, evaluation=best episode2013.12 | — | 21 | |
| HNeat Bestpolicy=deterministic2013.12 | — | 19 | |
| HNeat Pixelpolicy=deterministic, input=8 color channel representation2013.12 | — | -16 |