Index cannot be written with scipy.io.savemat

[cpcloud]

because i live dangerously and run master pretty much all the time i found this in some code of mine that saves indices to matlab files, the solution is to call .values, but it would be nice if we could not break this.

Here's a simple test:

import pandas as pd
import numpy as np
from scipy.io import savemat

idx = pd.Index(np.arange(10))

savemat('test.mat', {'idx': idx}, oned_as='column')

i'll look into fixing this

[jreback]

hmm, what does this call, Index is pretty ndarray-like

[cpcloud]

not sure, haven't looked into it yet

[cpcloud]

wow savemat is ... 😡

it looks at the __dict__ of the object if it's not a mapping but has that attribute and takes the attributes that don't start with _, in the case of this index it ends up being name which is None and it gets turned into a recarray with the name value pairs.

array([(None,)],
      dtype=[('name', 'O')])

.values it is

[jreback]

in theory this could be solved by if np.ndarray allowed isinstance to return True when using a MetaClass (but this is heavy magic and I could never get it too work).

[cpcloud]

yeah the __subclasshook__ stuff. i might chec kthat out but no guarantees

[jreback]

go for it. (FYI I tried with Series for a while, but I think numpy bypasses this somehow, could never get it to work)