Timeline for Why is SciPy acting very differently in IPython and Python?
Current License: CC BY-SA 3.0
28 events
when toggle format | what | by | license | comment | |
---|---|---|---|---|---|
Oct 12, 2021 at 6:25 | vote | accept | MWB | ||
Feb 10, 2014 at 23:22 | comment | added | Dschoni | I noticed, that dynamic cross-linking with openBLAS seems to be causing some problems when using with numpy - not sure if this is one of them. I ran into problems concerning core affinity when using numpy together with multiprocessing. These two cases MIGHT be somehow linked. | |
Feb 9, 2014 at 20:07 | comment | added | varantir | for me both ipython and python works. MaxB, if your problem is solved, you can mark it as solved. | |
Jan 2, 2014 at 3:47 | answer | added | SKV | timeline score: 1 | |
Dec 29, 2013 at 18:47 | comment | added | tsbertalan |
I made a virtual Ubuntu 12.04 64bit machine and installed the cited versions of Numpy, SciPy, Python, and IPython, and both versions of the script worked in both ipython and python (see https://gist.github.com/tsbertalan/8173209 ). However, after installing libopenblas-base , only the ipython one worked, as described.
|
|
Dec 27, 2013 at 10:24 | comment | added | MWB | @Shurane upon further investigation, the only solution I found was to not use these SciPy functions, and use the equivalent NumPy functions | |
Dec 23, 2013 at 21:06 | comment | added | Ehtesh Choudhury |
@MaxB might want to mark that as an answer. So do you mean you used something along the lines of apt-get remove python-numpy python-scipy ipython && pip install scipy numpy ipython ?
|
|
Dec 23, 2013 at 8:24 | audit | Close votes | |||
Dec 23, 2013 at 8:24 | |||||
Dec 16, 2013 at 23:38 | audit | Close votes | |||
Dec 16, 2013 at 23:54 | |||||
Dec 16, 2013 at 15:36 | audit | Reopen votes | |||
Dec 16, 2013 at 15:37 | |||||
Dec 15, 2013 at 19:19 | comment | added | ljk07 |
I found that adding this import to the plain python version of the code made the results match: from numpy.dual import register_func .
|
|
Dec 15, 2013 at 19:19 | comment | added | wim | No repro for me on 13.10 | |
Dec 14, 2013 at 5:17 | audit | Reopen votes | |||
Dec 14, 2013 at 5:17 | |||||
Dec 11, 2013 at 7:41 | comment | added | MWB | @Dougal I gave in to peer pressure to upgrade iPython, NumPy and SciPy, so I wiped them, and installed the latest versions from source instead of Ubuntu 12.04. I don't seem have this problem anymore. Thanks to everyone for trying to help! | |
Dec 11, 2013 at 7:32 | comment | added | Danica |
@MaxB Can you check for differences in environment variables? e.g. print os.environ from both. It seems like you're using openblas, and it might have different environment variables set to control things like multithreading.
|
|
Dec 11, 2013 at 1:10 | comment | added | MWB | @jtaylor Please see "edit 2". | |
Dec 11, 2013 at 1:06 | history | edited | MWB | CC BY-SA 3.0 |
added 186 characters in body
|
Dec 11, 2013 at 1:04 | comment | added | MWB |
@Dougal I tried it with flapack.so (I don't have _flapack.so in that directory). The results are identical from the shell and IPython, except for the numbers in the parentheses - I verified it in vimdiff .
|
|
Dec 10, 2013 at 20:37 | comment | added | jtaylor |
what is the output of: np.geterr() ? importing pandas, e.g. in your ipython profile, changes the error handling to ignore. which blas are you using (ls -l /etc/alternatives/libblas.so.3)
|
|
Dec 10, 2013 at 18:59 | comment | added | Danica |
What's the output of ldd /usr/lib/python2.7/dist-packages/scipy/linalg/_flapack.so ? Is it different if you run it from inside ipython (prefix with ! )?
|
|
Dec 10, 2013 at 18:41 | history | edited | MWB | CC BY-SA 3.0 |
added 473 characters in body
|
Dec 10, 2013 at 15:08 | comment | added | Warren Weckesser | @alko: Those are the versions in the Ubuntu 12.04 repository. | |
Dec 10, 2013 at 14:15 | comment | added | ev-br |
you do not seed the random number generator (np.seed(1234) ).
|
|
Dec 10, 2013 at 13:17 | comment | added | Deelaka | oops guess the output is random BTW it runs in less than a fraction of a second on numpy 1.7.1 and scipy 0.12.0 | |
Dec 10, 2013 at 13:13 | comment | added | alko | By the way, why numpy and scipy are so outdated? | |
Dec 10, 2013 at 12:55 | comment | added | Deelaka |
BTW if this (-0.0227526994852+0j) is the expected output it runs fine on my windows 32-bit machine with python 2.7, can you confirm this is the expected output??
|
|
Dec 10, 2013 at 12:53 | comment | added | user2357112 |
Dunno if this has anything to do with your problem (since I'm clueless about what the real problem is), but why do you have that np.array call converting the random sample to float32 ?
|
|
Dec 10, 2013 at 12:48 | history | asked | MWB | CC BY-SA 3.0 |