Programming Machine Learning

Hands On: Network Adventures

We didn’t write any code in this chapter, so why don’t you try writing it yourself? Here’s a task for you: set up a timer—maybe thirty minutes, or one hour, depending on your Python coding experience. Then take the perceptron-based code of the MNIST reader from Part I of this book, and try to convert it to the neural network that we designed in this chapter. Take as many shortcuts as you need. For example, feel free to ignore the softmax, and just use a second sigmoid instead. Stop when the timer rings.

Just to be clear, nobody expects that you’ll complete the job in one hour. In fact, you might have no code at all when the timer rings. That’s okay! The point of this exercise is not to code a neural network. The point ...