BUG: SARIMAX does not accept NumPy integer periodicty

[hylkedonker]

Passing a NumPy integer (instead of a Python integer) to the periodicty in the seasonal_order argument of SARIMAX raises the exception:

TypeError: object of type 'numpy.int64' has no len()

The following minimal code snippet reproduces the problem:

from numpy import *
from statsmodels.tsa.statespace.sarimax import SARIMAX

t = linspace(0, 8.0*pi, num=40 + 1)
y = sin(t) + random.normal(scale=0.5, size=len(t))
model = SARIMAX(
    y,
    seasonal_order=(0, 1, 0, int64(10))).fit()

The problem is in fact related to isinstance(int64(), int) being False, see also numpy/numpy#2951.
I am ussing statsmodels version 0.9.0, but looking at the master branch https://github.com/statsmodels/statsmodels/blob/master/statsmodels/tsa/statespace/tools.py#L181 I expect that this problem is not limited to my version only.
As a quick fix in my own code I do:

seasonal_order = tuple(int(i) for i in seasonal_order)

but it would be desirable if SARIMAX also works with NumPy integers.

[ChadFulton]

Thanks for the report! This shouldn’t be too hard to fix, I think. I never ran into this myself - is there a particular use case where Numpy ints are used?

…

[bashtage]

NumPy ints come up all the tie when automating runs where parameterizations are stored as NumPy arrays. Best practice is to sanitize using int(x) (or long, it is can be big and you care about Python 2.x) at the user boundary.

[bashtage]

You could also check isinstance(x, (int, np.integer)) to allow any numpy integer type through too.

[ChadFulton]

NumPy ints come up all the tie when automating runs where parameterizations are stored as NumPy arrays. Best practice is to sanitize using int(x) (or long, it is can be big and you care about Python 2.x) at the user boundary.

Thanks!