WARN: possible windows x64 issue with older numpy MKL

[jreback]

I happened to have an older verion of numpy 1.6.2 MKL installed on a windows machines

very intermittently (I think you hae to be using a really big frame)
sometimes I got a:

OMP : Error # 134 Cannot set thread affinity mask

which I think is caused by numexpr operations

see this

http://software.intel.com/en-us/forums/topic/279406

I am not sure its a real problem, because it only affects some versions of MKL, upgrading numpy (or even using a newer version of 1.6.2) seems to fix it..

also can be programatically disabled by

pandas.core.expressions.set_use_numexpr(False)

just a FYI

[davidljung]

I'm seeing the same error using Pandas 0.13.1 with Numpy 1.8.0 in Windows 7. It also occurs with expressions such as:
df = df.loc[(df['timediff'] > -20) & (df['timediff'] < 20),:]
and
df['GP'] = (df['Price'] - df['Cost'] - (df['Discount'] / df['Qty'])) * df['Qty']

They pop-up and empty shell window titles pythonw.exe and the error text given is:

OMP: Error #134: Cannot set thread affinity mask.
OMP: System error #87: The parameter is incorrect.

Unfortunately, there is no expressions module in pandas.core in this version, so I haven't figured out how to work around it yet.

(I think I'll open a new bug report, since this indicates an older issue and a build problem, which it may not be)

[jreback]

pls show the frame exactly as constructed. it appears you are trying to compare timedeltas with integers which is not implemented exactly right ATM IIRC.

I think the expression is passed to numexpr which is an open bug (it shouldn't be ), and instead evaluated directly (which does work)

[jreback]

I don't recall if timedelta ops were fixed in this or not #4924

[davidljung]

pd.computation.expressions.set_use_numexpr(False)

Does avoid the issue.
The timediff column is float64 type. I tried it with > -20.0 and < 20.0 and the error still occurs.

[jreback]

pls show how you constructed this frame

[JohnNapier]

Please find here (#9328) code to reproduce this issue.

[jreback]

closing as non-supported numpy version