I made a model for clustering and it's encoded dimension is about 3000. To check if the autoencoder is well established, I draw a 2d_pca plot and 3d_pca and the plots look nice. My question is that, what is general way to cluster with this encoded features?
I think about some options:
First: to use all encoded features.
Second: to use all encoded pca features.
Third: to use some encoded pca features explaining almost 70% variance.
I think usual papers don't refer to it.
