I have a dataset containing both categorical and numerical features. I am trying to work with Autoencoders for feature selection, so the first thing I do is to normalise the numerical features. For categorical features, I would like to apply categorical embedding, instead of Label Encoder or One Hot Encoder.
As the categorical embedding method will create N different variables to represent each categorical feature, the Autoencoder will choose one or many of those variables as selected features.
Thus, how could be the best way to face this situation?