Hands-On Reinforcement Learning for Games

Computational graphs with tensors

At the core of all DL frameworks is the concept of a tensor or what we often think of as a multidimensional array or matrix. The computational graphs we construct will work on tensors using a variety of operations to linearly transform the inputs into final outputs. You can think of this as a kind of flow, and hence the reason TensorFlow has the name it does. In the following exercise, we are going to construct a two-layer DL network using a computation PyTorch graph and then train the network:

The concepts here assume an understanding of linear algebra and matrix multiplication and systems of linear equations. As such, it is recommended that any readers lacking in these skills or are in need of a quick refresher ...