Design rationale/documentation

[mloning]

Do you have any documentation on the new API design and lessons from the forecast package?

[mitchelloharawild]

Not at the moment, but it would be a good blog post!

[mitchelloharawild]

The main points (some of which I've written about in vignettes/blogs) would be:

• the limitations of ts and the need for tsibble (sub-daily data, multiple series, leveraging existing user skills)
• the importance of consistent outputs (in forecast, some models gave forecasts while most gave models), and consistent accessors to them (tidy(), glance(), components(), fitted(), etc.)
• emphasising distributions over point forecasts and intervals
• the use of model definitions - useful for future interoperability with tidymodels, and are useful for interfaces which use models (forecast combinations)
• extensible by design
• areas of improvement in the current design

Do you have something similar for sktime?

[mloning]

Hi @mitchelloharawild, thanks, looks good, where can I find the vignettes/blog posts you mention?

For sktime, we've got a preprint and a more extensive design proposal in our wiki, but a lot of it is still work in progress.

I want to include a reference to fable, what's the best way to cite fable?

[mitchelloharawild]

We'll be writing a paper about fable and its design but we haven't started on this yet.

The fable introduction blog gives an overview of the fable interface: https://www.mitchelloharawild.com/blog/fable/

The fpp3 textbook also details a workflow for forecasting: https://otexts.com/fpp3/a-tidy-forecasting-workflow.html

My comparisons between ts and tsibble must have been scraped, I recall writing about it in the fable introduction blog but likely removed it for straying too far from the topic.

The current best reference for fable is:

citation("fable")
#> 
#> To cite package 'fable' in publications use:
#> 
#>   Mitchell O'Hara-Wild, Rob Hyndman and Earo Wang (2020). fable:
#>   Forecasting Models for Tidy Time Series. R package version 0.1.2.
#>   https://CRAN.R-project.org/package=fable
#> 
#> A BibTeX entry for LaTeX users is
#> 
#>   @Manual{,
#>     title = {fable: Forecasting Models for Tidy Time Series},
#>     author = {Mitchell O'Hara-Wild and Rob Hyndman and Earo Wang},
#>     year = {2020},
#>     note = {R package version 0.1.2},
#>     url = {https://CRAN.R-project.org/package=fable},
#>   }

^{Created on 2020-06-22 by the reprex package (v0.3.0)}