numpy.dtype size changed runtime warning on XP

[tgefell]

I've installed numpy from numpy-1.5.1-win32-superpack-python2.6.exe
and pandas from pandas-0.2.win32-py2.6.exe.

When I import * from pandas I get the following warnings:

Warning (from warnings module):
File "C:\Program Files\Python26\lib\site-packages\pandas\core\index.py", line 7
from pandas.lib.tseries import map_indices
RuntimeWarning: numpy.dtype size changed, may indicate binary incompatibility

Warning (from warnings module):
File "C:\Program Files\Python26\lib\site-packages\pandas\core\index.py", line 7
from pandas.lib.tseries import map_indices
RuntimeWarning: numpy.flatiter size changed, may indicate binary incompatibility

I've had no issues yet using pandas interactively in the window (idle), but if I have embedded the import in a module then I am not able to execute any of the code from the module.

Thoughts?

[wesm]

That pandas binary was built against NumPy 1.3.0 which is not ABI-compatible with the latest NumPy-- are you equipped to compile from source? I should probably post a 0.3 beta anyway on PyPI.

[tgefell]

I'm rather a nubi at all of this so I'm not sure I should try compiling from source. It took me about 3 days to get registry, env var's etc. set up to use at all. Should I revert to numpy 1.3 or should I try your 0.3? I'm not doing anything that's production sensitive, so I'd be game to try the 0.3.

[wesm]

The github HEAD is well-tested but not yet ready for release until I do a few more things. I'll build and post 0.3 beta binaries later today-- stick with NumPy 1.5.1.

BTW I'd recommend checking out the Enthought Python Distribution as it comes set up to compile things from source with no hassle.

[tgefell]

Very good. Thanks Wes. I'm just trying to get enough traction learning Python/ now pandas and actually get something done at the same time.

[wesm]

I posted binaries for Python 2.5-2.7 built against latest NumPy-- let me know if you have any trouble!!

http://pypi.python.org/pypi?:action=display&name=pandas&version=0.3.0.beta

[tgefell]

Wes, I'm new enough that I've never moved up a version on anything python-related. Do I need to uninstall my earlier version of pandas before running the new installation .exe? Will this new installer automatically fix up any references in my path variable and my registry?

[wesm]

If you run the new installer it will override your prior installation-- you can check that the right version is installed by:

import pandas
print pandas.version

[tgefell]

Wes,

I'm not sure what you need me to send you to help you debug. I did install 0.3 and it did load without the warnings. However, it also bombed when I tried to create a DataFrame. I cut and pasted the following, with some notes included:

I create a list: headers in row 1, dates in col 1, values in col's[1:]

t = gf.get_bmrk_asset_weights()

type (t)
<type 'list'>
len(t)
2037
t[:2]
[['Date', 'BA', 'CAT', 'CMI', 'CSX', 'DE', 'DHR', 'EMR', 'FDX', 'GD', 'GE', 'HON', 'ITW', 'LMT', 'MMM', 'NOC', 'NSC', 'PCAR', 'PCP', 'RTN', 'TYC', 'UNP', 'UTX', 'WM', 'XLI'], [datetime.date(1998, 11, 25), 0.062619577947945801, 0.027268106641251115, 0.0022265456759515082, 0.014247168087848643, 0.013195744171185141, 0.0095164984532162631, 0.043821780902251181, 0.014310929221289945, 0.011549577203586944, 0.46356668152426372, 0.038046569176042495, 0.025077286132099617, 0.032061521644136844, 0.051246232439461852, 0.0085445846506965115, 0.018797841874526829, 0.0056623434171960965, 0.0016919000630261892, 0.029560646985546359, 0.031922116996210739, 0.018288231848329117, 0.03828907124638279, 0.03848904369755457, 0.0]]

I split out the dates as my DataFrame index:

i = [x[0] for x in t[1:]]
len(i)
2036
i[:2]
[datetime.date(1998, 11, 25), datetime.date(1998, 11, 27)]

I split out cols[1:] as a dictionary

d = dict([(t[0][j+1], [x[j+1] for x in t[1:]]) for j in range(len(t[0])-1)])
d.keys()
['RTN', 'BA', 'CMI', 'PCP', 'DE', 'CAT', 'EMR', 'FDX', 'NOC', 'PCAR', 'XLI', 'NSC', 'WM', 'UNP', 'HON', 'GE', 'GD', 'CSX', 'UTX', 'DHR', 'LMT', 'TYC', 'MMM', 'ITW']
len(d)
24
len(d["BA"])
2036

The following line dies with a Microsoft Window saying "pythonw.exe has encountered a problem and needs to close. We are sorry for the inconvenience.":
df = DataFrame(d, index=i)

This worked with the pandas 0.2 that I had had installed.

[wesm]

Can you please try running

pandas.test() and copy-paste the output-- you may need to install nose (!)

I will take a look shortly and let you know

[tgefell]

I did install nose.

This doesn't look very helpful:

from pandas import *
pandas.test()
Traceback (most recent call last):
File "<pyshell#2>", line 1, in
pandas.test()
NameError: name 'pandas' is not defined

I don't get it. It clearly knows about pandas. It loaded.

[wesm]

sorry:

import pandas
pandas.test()

[tgefell]

Same error window/message after this:

import pandas
pandas.test()
Running unit tests for pandas
NumPy version 1.5.1
NumPy is installed in C:\Program Files\Python26\lib\site-packages\numpy
Python version 2.6.6 (r266:84297, Aug 24 2010, 18:46:32) [MSC v.1500 32 bit (Intel)]
nose version 0.11.3

[wesm]

let's discuss offline on gmail, will be easier (i'm wesmckinn at gmail)