tsibble + fasster + hilo

[earowang]

• [ ] fasster API with tsibble
• [ ] fasster model object
• [ ] forecast object integrated with hilo
• [ ] update readme, using tsibble::pedestrian

[mitchelloharawild]

Perhaps a good place to start would be a lengthy discussion about how forecast objects should be structured for forecast 2.0 😄

[mitchelloharawild]

Adding to this, a vignette would be useful for fasster, once the API is confirmed.

[earowang]

Basic idea:

[earowang]

if a tsibble is a univariate time series, fasster returns a fasster model object. if it is multiple time series, fasster could return a table of model objects for each key. forecast() will gain a tbl_model and fasster (already implemented). Inside the forecast.tbl_model, it will call the forecast.fasster for each key. forcast.tbl_model returns a tbl_forecast.

tsibble will have support for tbl_model and tbl_forecast (a collection of models), but no intention to support a single model object. I think this way make the time series forecasting extensible for other developers.