About the loss function of the model

[Foxyrep]

Hello, at first thank you for posting this work, and it helps me a lot on my comprehension about TCN and AE! This is what I'm trying to learn these days!
Well there is a question I'm confused about. It's about the loss function in the model compiling.
In the .ipydb file, In[17], Line 173, well since we have set the latent loss, why don't we use the latent loss in the model compiling?
Hope you could help me on this question, thx!!

[tvhahn]

Thanks for the compliment! I'm glad it helped.

We use the latent_loss (line 129 in [17]) by using the Keras add_loss method (line 138 in [17]). There is a good explanation of why Keras has this "add_loss" method in this stackoverflow post: https://stackoverflow.com/a/52683522/9214620.

Aurélien Geron has an easier to understand implementation of a VAE in his Jupyter notebook. It's worth looking at. His explanation in the book is good too. https://github.com/ageron/handson-ml2/blob/master/17_autoencoders_and_gans.ipynb

Let me know if you have any other questions. Cheers!

[Foxyrep]

Grateful for your so quick reply! I'll go through the references you advised later, thx a lot!