Implementation

Now it's time for the implementation! The first thing you need is a professional toolkit, because you're not going to build an artificial brain with simple Python libraries. What you need is an advanced framework, which allows fast computation for the training of neural networks.

Today, the best frameworks to build and train AIs are TensorFlow (by Google) and PyTorch (by Facebook). How should you choose between the two? They're both great to work with and equally powerful. They both have dynamic graphs, which allow the fast computation of the gradients of complex functions needed to train the model during backpropagation with mini-batch gradient descent. Really, it doesn't matter which framework you choose; both work very well for our self-driving car. As far as I'm concerned, I have slightly more experience with PyTorch, so I'm going to opt for PyTorch and that's how the example in this chapter will continue to...