Interface features both parts of the ELBO-loss (likelihood part and performance part)

[filljonas]

For my recent work about Bayesian Neural Networks for Time Series Forecasting in Hydrology I needed a slight modification of the BLiTZ-library where the function _sampleelbo in _variationalestimator.py does not only output the total loss, but the likelihood part and performance part individually. In addition, my version outputs the model predictions.

As I think that this could be sensible addition for public use of the library, I would like to apply for a change here.

Best regards, Jonas Fill

[piEsposito]

Hi, and thank you for contributing to BLiTZ. I'm gonna run the tests and review the PR.

Tks, -Pi

[piEsposito]

The tests did not pass as you changed the return value of a important function of @variational_estimator.

I've presented some approaches you can go to to solve this issue, and you can always create a new function return the values as you wish without changing this function that is a central part of BLiTZ's API.

Tks!

[filljonas]

Hi, I restored the original function within @variational_estimator and added my own function where I changed the return type. Please let me know if there is anything else that I need to adjust.

Best, Jonas

[piEsposito]

Gonna run the tests and let you know. Thanks again :D

[piEsposito]

Now it works. I'm merging it, thou it would be nice if you could add the unit tests for this function on other PR.