[Numpy-discussion] How to debugging python crash in ipython notebook

[Åsmund Hjulstad]

Hello all,

I am struggling with a python crash of an ipython notebook (kernel) that I
do not know how to debug.

If I run this:  (valgt is a pandas dataframe with 354 lines and 14
numerical columns)

sns.pairplot(valgt)
plt.savefig('parvis.png', dpi=600)

in the same cell, the kernel consistently crashes. If I run in two separate
cells, or exporting to a Python script and running there (changing the
inline render to qt), everything is OK.


This is win64 on Win7, both the free build from conda, the accelerate build
from conda, and also a version I have built myself. (VS2010 and Intel
Fortran XE 2015)

The following is the package list from the 'vanilla' conda:

# packages in environment [...]:
#
dateutil                  2.1                      py34_2
ipython                   2.3.1                    py34_0
jinja2                    2.7.3                    py34_1
markupsafe                0.23                     py34_0
matplotlib                1.4.2                np19py34_0
numexpr                   2.3.1                np19py34_0
numpy                     1.9.1                    py34_0
pandas                    0.15.2               np19py34_0
pip                       6.0.6                    py34_0
pyparsing                 2.0.1                    py34_0
pyqt                      4.10.4                   py34_0
pyreadline                2.0                      py34_0
python                    3.4.2                         1
pytz                      2014.9                   py34_0
pywin32                   219                      py34_0
pyzmq                     14.5.0                   py34_0
rpy2                      2.5.5                     <pip>
scipy                     0.15.1               np19py34_0
seaborn                   0.5.1                np19py34_0
setuptools                12.0.5                   py34_0
six                       1.9.0                    py34_0
tornado                   4.0.2                    py34_0

rpy2 is from the Gohlke python collection.


The home-built version is a python 3.4.3rc1, with

certifi (14.5.14)
Cython (0.21.2)
ipython (2.4.1)
Jinja2 (2.7.3)
jsonschema (2.4.0)
MarkupSafe (0.23)
matplotlib (1.4.2)
mistune (0.5)
nose (1.3.4)
numexpr (2.4.1.dev0)
numpy (1.10.0.dev0+98a8fe3)
pandas (0.15.2)
pip (6.0.8)
pyparsing (2.0.3)
pyreadline (2.0)
PySide (1.2.2)
python-dateutil (2.4.0)
pytz (2014.10)
pywin32 (219)
pyzmq (14.5.0)
rpy2 (2.5.6)
scipy (0.15.1)
seaborn (0.6.dev0)
setuptools (12.0.5)
six (1.9.0)
tornado (4.1)
wheel (0.24.0)


They crash in the same way, though, the same applies to a test I did with
ipython 3


I am wondering how important it is to have everything built with the same
compiler and MKL library. Does all packages that use numpy need to be built
against the specific numpy+mkl version? For example rpy2 (which is not used
near the cells where it crashes), should I always rebuild this as well if I
build differently, or...?

And, how can I generate debug information from the kernel that ipython
notebook starts?

Any pointers are most appreciated.




-- 
mvh,
Åsmund Hjulstad
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/numpy-discussion/attachments/20150210/63172d77/attachment.html>