October 2019
Intermediate to advanced
340 pages
8h 39m
English
The observation in Pong is much more complicated than the environments we have worked with so far in this chapter. It is a three-channel image of 210 * 160 screen size. So, we first transform it into a grayscale image, downsize it to 84 * 84, and then flatten it so that it can be fed into the fully connected neural network. As we have inputs of around 6,000 dimensions, we use two hidden layers to accommodate the complexity.
Read now
Unlock full access