Build a Neural Network with Keras

Python boasts a few large ML libraries such as PyTorch and TensorFlow. They are complex, and they tend to evolve fast. We will not use them in this course.

Instead, we will use a slimmer and more stable library named Keras. Keras is a thin layer that sits on top of the “big” libraries and hides them behind a nice, clean programming interface. Keras started out as a tool for prototyping, but it quickly became one of the most popular ways to build neural networks in Python — it was even adopted by TensorFlow as an official interface. With Keras, we can use the heavyweight libraries without getting bogged down by their complexities.

A plan and a piece of code

Let’s build a neural network for the Echidna dataset. We can use the MNIST network that we designed back in Designing the Network as a starting point.

As a reminder, $M$ is the number of examples in the training set, and $784$ is the number of input variables in MNIST. Here is how we can modify this network to learn the Echidna dataset instead:

1. The Echidna dataset has two input variables, so we only need two input

...