- Implementing Deep Q-Learning agent to play Pong.
- Refer to the following in order to understand the implementation:
Here is the final agent, which seems to lose the first point in order to put the other agent in worse position:
However, 20 episodes before at 500 - the agent competes with the game without losing a single point.
As shown by the graph below, the agent starts very slowly with a reward of -20 and eventually gets reasonably consistent rewards of 20: