Fixing Sarima with exogenous variables

[Axemen]

When using SARIMA it would not pass exogenous variables to the underlying statsmodels.tsa.statespace.sarimax.SARIMAXResults function.

Added a check to ensure the user provides the exogenous variable needed for forecasting if they provided an exogenous variable in the SARIMAParams initialization.

Added test cast for exogenous variable with a daily frequency

I'm not exactly sure how to update the documentation since this added a new parameter to the kats.models.SARIMA.predict function and would like further guidance on how to do so.

Kats

Thank you for contributing! Please take a moment to review our contributing guidelines
to make the process easy and effective for everyone involved.

• Fork the repo and create your branch from master.
• If you've added code that should be tested, add tests.
• If you've changed APIs, update the documentation.
• Ensure the test suite passes.
• Make sure your code lints.
• If you haven't already, complete the Contributor License Agreement ("CLA").

Contributor License Agreement ("CLA")
In order to accept your pull request, we need you to submit a CLA. You only need to do this once to work on any of Facebook's open source projects.

Complete your CLA here: https://code.facebook.com/cla

Ver: Kats v1.02

[kirkdotcam]

We also weren't sure if we should allow exog to be passed as a named parameter or a kwarg, so here is the code for the kwarg version if that's preferred:

if (self.params.exog is not None) and (kwargs.get('exog') is None):
    msg = "SARIMA model was initialized with exogenous variables. Exogenous variables must be used to predict. use `exog=`"
    logging.error(msg)
    raise ValueError(msg)
fcst = self.model.get_forecast(steps, exog=kwargs.get('exog'))

[facebook-github-bot]

@iamxiaodong has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[facebook-github-bot]

@jeffhandl merged this pull request in 90b69ed.