Programming Machine Learning

Hands On: Hyperparameters Galore

A CNN has more hyperparameters than you can shake a stick at. On top of the usual hyperparameters of fully connected layers, you also have to pick the number and size of filters in each layer. In case all those decisions aren’t feeling overwhelming yet, also consider the additional parameters I mentioned in ​Convolutions Can Get Complicated​.

Your mission is to experiment with the hyperparameters of the CNN we built in this chapter. Tune those that you already know about, and maybe search online for the meaning of those we skipped. You probably won’t be able to do very fine hyperparameters tuning, because that would require training the CNN for a long time—but you’ll get to understand those options a little bit ...