Extract embedding of latent space in TemporalFusionTransformer

[Mpaperlee]

[DOC] I am using the pytorch_forecasting to create a TemporalFusionTransformer model. After train the model, I want to extract the data embedding in latent space, than use it to clustering. But I cannot find out the solution. How should I do

[benHeid]

If I remember correctly, we do not return the output of the encoder currently. So would probably need some changes to the code to enable it. There are two possibilities to enable it. Either by extracting the relevant code into a new function or by adding the encoded data as additional field to the network output.

@fkiraly @XinyuWuu @julian-fong @pranavvp16 any opinions about the two solutions?