Unable to handle "nans"

[Jakobhenningjensen]

Description

When the dataset contains "nans", it seems to fail. When using auto.arima in R, these are handled/omitted

Steps/Code to Reproduce

Expected Results

Ommit nans

Actual Results

Raises an "ValueError"

Versions

Windows-7-6.1.7601-SP1
Python 3.6.7 |Anaconda custom (64-bit)| (default, Oct 28 2018, 19:44:12) [MSC v.1915 64 bit (AMD64)]
pmdarima 1.0.0
NumPy 1.15.4
SciPy 1.1.0
Scikit-Learn 0.20.0
Statsmodels 0.9.0

[tgsmith61591]

First off, thanks for the well-formatted issue. I understand the frustration around NaNs; R handles them implicitly (albeit in an almost black-box fashion) while most python libraries make you, the programmer, address them first.

The reason for the error is due to the call to scikit-learn's check_array which does not tolerate NaNs. Best strategy for now is to impute them yourself, and if scikit makes the move towards handling NaNs it might make sense for us to as well.

[tgsmith61591]

Closing since this is currently a "wontfix." But will leave the backlog tag in case scikit decides to support this behavior later.