concept MLP in category python

This is an excerpt from Manning's book Math for Programmers: 3D graphics, machine learning, and simulations with Python MEAP V11. Login to get full access to this book.

Figure 16.11: Two layers of activations for our multilayer perceptron (MLP) calculated.

If we want our class to represent an MLP, we need to specify how many layers we want and how many neurons we want per layer. To initialize our MLP with the structure we want, our constructor can take a list of numbers, representing the number of neurons in each layer.

The data we need to evaluate the MLP are the weights and biases for every layer after the input layer. As we just covered, we can store the weights as a matrix (a NumPy array) and the biases as a vector (also a NumPy array). To start, we can use random values for all of the weights and biases and then when we train the network, we can gradually replace these values with more meaningful ones.