WIP: implement quantile loss function

[trdvangraft]

Add quantile loss to fastai, solves #3279

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[ReinierKoops]

Every file seems to have been edited, maybe a new pull request with only the actual feature change?

[ReinierKoops]

Ah, now I see. The whole library had to be rebuild because of the addition of your one feature.

[hamelsmu]

@ReinierKoops I converted this to a draft PR while you sort the conflicts (and also b/c the PR title says WIP)

[jph00]

@trdvangraft please let me know when this is ready to review

[tcapelle]

Would be cool to do something like this: https://www.angioi.com/time-series-encoder-decoder-tensorflow/ using the new GaussianLoss in pytorch 1.8, basically, instead of doing a regression, you predict mean and std and fit a normal distribution that minimizes NLL. I can help you with this if you want.

[trdvangraft]

Hey sorry for the long radio silence, I got tangled up in some other work. I think we can package this MR together @tcapelle to add some more loss functions. But first we will have to figure out how to get rid off the notebook duplication and get it updated with master again.

[tcapelle]

Just create a new PR