Why does scikit neural network reshape my input array automatically?

Question

When I pass my training data to the scikit neural network it is a NumPy array of all my 24*24 image matrices. I check and this is the shape of the data: (3237, 24, 24) - 3237 24*24 images.

However, once I feed it into the neural network, I get this warning:

WARNING:sknn: - Reshaping input array from (3237, 24, 24) to (2589, 24, 24, 1).

The program still runs, I just don't understand why it is reshaping my array.

Answer 1

Looking at the source, scikit will reshape input arrays if the X (input samples) you pass in when initializing is a different size from creating/splitting the dataset in the MLP backend. This is to reduce overfitting by training and validating on the same samples.