Hands-On Machine Learning for Algorithmic Trading

A simple feedforward network architecture

A feedforward neural network consists of several layers, each of which receives input data and produces an output. The chain of transformations starts with the input layer, that passes the source data to one or several internal or hidden layers, and the output layer that computes a result that can be compared to the outcome or label of the ML problem.

The hidden and output layers consist of nodes or neurons, each of which connects to some or all nodes of the previous layer, where the latter is called a fully-connected or dense layer. The network architecture can be summarized by the depth of the neural network, measured by the number of hidden layers (hence deep learning), and the width, or the number ...