Train on custom data-set with more than one class

How to change the code to make it work on custom dataset with more than one class.


I see that it's done for MNIST data-set. So I want to do the same for my dataset which contains data of airplanes, cars, bikes,pants etc...

Is it through y_dim?

And also where are I need to make changes to make it work?

Start by tracing how the MNIST dataset is loaded and how y_dim is used in the training path. Determine the changes needed for a custom dataset with multiple classes such as airplanes, cars, bikes, and pants, and verify that training completes with the resulting class configuration.

Written by the indexing model from the issue text.