How to track KL divergence and reconstruction loss separately during training?

matsengrp / vampire

🧛 Deep generative models for TCR sequences 🧛

Apache License 2.0

16 stars 4 forks source link

How to track KL divergence and reconstruction loss separately during training? #122

Closed dandanpeng closed 4 years ago

dandanpeng commented 4 years ago

This may be a stupid question. But I cannot figure out how to extract KL divergence and reconstruction loss from the multiple outputs model? If it's a single output model, I think 'add_loss' can solve my question.

krdav commented 4 years ago

Hi Dandan, Can you provide a bit of context? How are you running it using the scons workflow or directly working with the .py scripts? Some code showing what you mean would also be helpful.

dandanpeng commented 4 years ago

Hi Krdav, I wanted to edit the py. scripts and track the reconstruction error and KL divergence separately during training. Now I have solved it. This is basically a Keras related problem rather than Vampire-related. Anyway, thanks for the reply : )