Reducing StatsForecast Size

[kvnkho]

There are some issues saying that the size of StatsForecast is too inconvenient for AWS Lambda. See this and this. The package sits at 700 MB.

We can check the size of the dependencies below. Note that Spark, Dask, and Ray are dev dependencies. Plotly (and Dash) are core dependencies.

This pull request removes plotly as a core dependency in favor of matplotlib being the default.

The second thing is that the build includes all source notebooks (docs and source). The docs folder is around 70-75MB. Removing the docs will also save a lot of space, especially because the docs contain plotly images.

Note:

I had to change the test to pass. It has to do with Plotly and Matplotlib not having 1:1 parity with behavior. It's somewhat ok for now because the syntax will slightly differ, but users are unlikely to use both anyway. I can open another issue for that.

EDIT: Docs are actually not included in the build already because there is no init.py . Action files are but they are so small anyway it doesn't matter.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[jmoralez]

LGTM. I'm adding the breaking label because if someone is relying on statsforecast installing plotly and having it as the default backend this will require them to add it as a dependency and setting the engine argument.

[kvnkho]

LGTM. I'm adding the breaking label because if someone is relying on statsforecast installing plotly and having it as the default backend this will require them to add it as a dependency and setting the engine argument.

Yep that's right, unfortunately.

[DaveParr]

Looking forward to using this, thanks for the effort. 👍🏻

[nelsoncardenas]

Awesome job!