Sorry for the update: but now one version of the job (same size, but different
data) still failed with the OS failing to provide memory, even with n_jobs=8.
The maximum RAM the job took was 119705224K, while it was allocated 240 GB.
I am not sure how to make this reproducible for a bug report without sharing
the data.
On Fri, Aug 22, 2014 at 8:54 AM, László Sándor <sand...@gmail.com> wrote:
> Oh, Mailbox failed with the attachment, sorry, here it is.
> On Fri, Aug 22, 2014 at 8:43 AM, László Sándor <sand...@gmail.com> wrote:
>> Sorry, Olivier, I am not sure I follow. In any case, I am happy to update
>> skl, I was only using the latest 64-bit Linux from the conda distro (though
>> using MKL wanted to downgrade me to 0.14, I am pretty sure I precluded
>> that).
>>
>> That said, maybe it’s good I did not file a proper bug report: The job
>> completes with specifying n_jobs = 8 [the Windows test in July was an
>> 8-core server], while n_jobs = -1 might have meant that the 64-cores it was
>> supposed to find blew the job out of memory. I am still not sure why this
>> happened when the job currently occupied 120 GB out of 240, but this might
>> not be a scikit-bug after all.
>>
>>
>> On Fri, Aug 22, 2014 at 8:37 AM, Olivier Grisel <olivier.gri...@ensta.org>
>> wrote:
>>
>>> Thread based enet CV has been merged after the 0.15 release.
>>>
>>
>>
------------------------------------------------------------------------------
Slashdot TV.
Video for Nerds. Stuff that matters.
http://tv.slashdot.org/
_______________________________________________
Scikit-learn-general mailing list
Scikit-learn-general@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
