Hands-On Machine Learning for Algorithmic Trading

How neural networks work

In this section, we will introduce a fundamental class of (artificial) neural networks that is based on the Multilayer Perceptron (MLP) and consists of one or more hidden layers that connect the input to the output layer.

This class of neural networks is also called feedforward neural networks because information only flows in one direction, from input to output. Hence, they can be represented as directed acyclic graphs. In contrast, in the next chapter, we cover RNNs, which include loops from the output back to the input to enable the neural network to keep track of or memorize past patterns and events.

We will first describe the architecture and how to implement it using NumPy. Then we will describe backpropagation ...