Affine Coupling Quick Question


I’m implementing Normalizing Flows for zero-shot learning.

I have a quick question about What is the purpose of param_dims in defining the hypernet?


cc @stefanwebb

param_dims defines the output shape of the conditioning network. For the AffineCoupling transform it has a single mean and log(variance) for each dimension that is modelled, and there are input_dim-split_dim dimensions