Artificial Intelligence By Example

Modifying the hyperparameters

The training was done with the same CNN_STRATEGY_MODEL.py program described in Chapter 9, Getting Your Neurons to Work, which was designed to be generalized for subsequent chapters.

Once the dataset of the previous section has been installed, there are only two hyperparameters to change in the following code snippet:

estep=5000 #5000 steps per epoch...ep=3 #epochs

This forces the program to train longer:

• Three epochs: ep=3
• 5,000 steps per epoch: estep=5000

As the code that follows shows, the hyperparameters have been modified; the program runs and the new model is saved with its graph flow structure and weights:

# serialize model to JSONmodel_json = classifier.to_json()with open(directory+"model/model.json", ...