While an agent trains, the experience buffer recycles old memories and replaces them with new ones. As we discussed, the purpose of this is to break any localized patterns or, essentially, situations where the agent just repeats itself. The downside of this, however, is that the agent may forget what the endgame is, which is what happened in the last example. We can simply fix this by increasing the size of the experience buffer, which we will do in the next exercise:
- Open Visual Studio Code or your favorite text editor.
- Locate the trainer_config.yaml file in the python folder and open it.
- Locate the configuration for the HallwayBrain, as shown in the following code:
HallwayBrain: use_recurrent: true sequence_length ...