Learning OpenCV 4 Computer Vision with Python 3 - Third Edition

Choosing the size of the hidden layer

There are no agreed-upon rules of thumb for choosing the size of the hidden layer; it must be chosen based on experimentation. For every real-world problem where you want to apply ANNs, you will need to train, test, and retrain your ANN until you find a number of hidden nodes that yield acceptable accuracy.

Of course, even when choosing a parameter's value by experimentation, you might wish for the experts to suggest a starting value, or a range of values, for your tests. Unfortunately, there is no expert consensus on these points either. Some experts offer rules of thumb based on the following broad suggestions (these should be taken with a grain of salt):

• If the input layer is large, the number of ...