Certain DataFrame cannot be pickled to string

[craustin]

import cPickle
import numpy as np
from pandas import Series
s = Series({'a' : None})
s['a'] = np.inf
cPickle.loads(cPickle.dumps(s))

ValueError: could not convert string to float

[craustin]

This is likely fixed by the float -> float64 fix (#510). I'll verify.

[craustin]

Confirmed.

[craustin]

Chris is getting this error elsewhere. Do you think there's anything we can do? This repros in Pandas 0.4.0 as well.

[wesm]

Please let me know when you verify the fix-- should be taken care of by the float/float64 fix perhaps?

[craustin]

I thought it might be fixed by that float fix, but no luck. I still get this error on a build from 2012-01-05.

[wesm]

Any chance of reproducing? And this is on win32? Or rather the above code snippet fails on windows?

[craustin]

I can repro this with the original code in the issue. This is on EPD 6.3-2 + Windows 64.

[wesm]

OK, I am able to reproduce on win64 + epd 6.3-2

[wesm]

This is caused by a bug in NumPy 1.4.0. I verified that it's fixed in NumPy 1.6.1:

import numpy as np
import cPickle
arr = np.array([np.inf], dtype='O')
cPickle.loads(cPickle.dumps(arr))

However! Pickling with protocol 1 or 2 (or cPickle.HIGHEST_PROTOCOL) works

cPickle.loads(cPickle.dumps(arr, protocol=cPickle.HIGHEST_PROTOCOL))

You should be doing that anyway-- performance is a lot better. Make sure any file handles are open in binary mode.