What is the difference between categorical crossentropy and sparse categorical crossentropy in keras? When these loss functions are suitable?

Question

I am facing some errors, while using these loss functions. so please explain me, when to use these loss functions and with the output layer units.

Answer 1

I don't know the exact difference between these loss functions. I think the documentation and source code should be checked. But both of these loss functions are suitable for classification models. you should notice two things:

1. Your outputs should be converted to categorical: first use Sckit-learn class LabelEncoder() to covert strings to integers, then use to_categorical() Keras method on your output to covert your integers to a one-hot encoding
2. You need to use a softmax layer as your last layer with the same size of your categories