Deep Learning Quick Reference

Input layer shape

Since we've already identified our inputs, we know that the input matrix will have a number of rows equal to the number of data elements/observations in our dataset and a number of columns equal to the number of variables/features. The shape of the input matrix then is (number of observations x 10 features). Rather than defining the exact number of records in our dataset or minibatch, TensorFlow and Keras allow us to use None as a placeholder when we define the number of elements in a dataset.

If you see a None dimension used in a Keras or TensorFlow model layer shape, it really means any, the dimension could take on any positive integer value.