[FEAT] Cross-temporal forecasting

[elephaint]

First stab at (generic) temporal reconciliation. I've tried out a lot of stuff, but it seems for the moment it's easiest to implement using the strategy set out below.

Temporal reconciliation strategy

• Aggregate temporally using aggregate_temporal, which is a minimal wrapper around aggregate, but in the temporal dimension, using a provided temporal spec
• Reconcile using our current reconcilers

Thoughts

• The simple implementation I'm currently suggesting becomes quite involved as it requires a lot of data wrangling and user interaction (see the added Example), so offering additional helper functions to users to get rid of that may be helpful
• It may be beneficial as part of this PR to define a generic HierarchicalForecast class, with a.o. a .fit-method that consumes df, a cross-sectional and/or temporal spec, and a set of models linked to the spec. That would allow a user to only have to define (i) their spec, (ii) their models, (iii) their data, and run fit / forecast / cross-validation on the HierarchicalForecast class. Having fiddled with this over the last days there are a number of issues with this, a.o. that we need to generalize over the provided models in .fit (e.g. allowing models from SF, MF, NF), thus essentially building a Nixtla-class. Maybe that's a stretch for now.

Open issues

• First example notebook doesn't work yet; main issue to solve is how the temporal unique ids can be matched back to the forecast df
• Unit tests are missing
• More examples (only temporal, temporal -> cross-sectional)

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB