2

I'm currently using ARIMA with pyramid, and when creating an ARIMA object using pyramid's ARIMA() I can sepcify an exogenous parameter, but when calling fit() I cannot specify an exogenous variable.

However with statsmodels, I saw that this is reversed. I cannot specify an exogenous parameter with ARIMA() yet I can specify one with fit().

If I would like to include an exogenous parameter for both ARIMA() and fit(), which one should I use?

amadzebra
  • 175
  • 2
  • 3
  • 9

1 Answers1

4

There are several differences between statsmodels' ARIMA class and pyramid's (recently renamed to pmdarima):

  • First of all, statsmodels' ARIMA class has no seasonal component. pmdarima's ARIMA class allows seasonality optionally. It's an all-in-one wrapper for the statsmodels ARMA, ARIMA and SARIMAX
  • Statsmodels takes data in the constructors, but pmdarima adheres to more of a scikit-learn signature, which accepts only model hyper-parameters in the signature and the data in the fit call.

It's important to note that when you pass exogenous to pmdarima's fit call, it is internally passing the exogenous array to the underlying ARIMA class.

tl;dr

When you use pmdarima, it is using statsmodels under the hood. Just pass the data to the fit method.

TayTay
  • 6,882
  • 4
  • 44
  • 65