Advanced Deep Learning with Python

Generating the training dataset

In this section, we'll look at how to generate a training dataset and load it as an instance of PyTorch's torch.utils.data.DataLoader class. We'll highlight the most relevant parts of the code, but the full source code is located at https://github.com/PacktPublishing/Advanced-Deep-Learning-with-Python/blob/master/Chapter11/imitation_learning/train.py.

We'll create the training dataset in several steps:

1. The read_data function reads imitation_learning/data/data.gzip in two numpy arrays: one for the game frames and the other for the keyboard combinations associated with them.
2. The environment accepts actions, composed of a three-element array, where the following are true:
   • The first element has a value in the ...