Update numpy to 1.10

[durack1]

They've just released a bugfix update to the 1.9.x branch: http://docs.scipy.org/doc/numpy/release.html

I'll update the sources and issue an pull request when I deal with the other issues #856, #847, #846 (which is going to need some testing), and #841 and #814

[durack1]

@doutriaux1 @aashish24 I suggest this be set against the 2.2 milestone

[durack1]

Source is updated at the ${LLNL_URL} this includes:
numpy-1.9.1.tar.gz

I'll tweak the MD5 code (78842b73560ec378142665e712ae4ad9) and patch numbers and submit a pull request

[doutriaux1]

I think we won't do this just before 2.1 release. I'm afraid something comes up.

[durack1]

Yep, 2.2 is fine with me.. I'd imagine the other updates will also be released (and I will have issued the pull request) well within the 2.2 timeframe..

[durack1]

@doutriaux1 another candidate for a move to 2.3 - and likely we'll just skip 1.9.1 and go straight to 1.9.2 when it's out (soon - 1.9.2rc1 is out)

[durack1]

@doutriaux1 can you add build and cdms2 labels please?

[doutriaux1]

please no issue if you have the PR ;)

[durack1]

You don't want both?

[doutriaux1]

well... usually you create an issue to remind to fix it and THEN issue the PR. Once the PR is in, there's no need for the issue any longer. Except maybe for the README file that @chaosphere2112 auto-generates.

[durack1]

That's exactly why I generated both.. First issue to outline problem (issue) and then the PR to "solve" it.. Having both means that it'd be easier for @chaosphere2112 to then just poll the closed issues directly rather than having to generate an issue from an orphan PR..

[durack1]

@dnadeau4 @doutriaux1 another candidate for 2.4 - and likely just skip 1.9.1 and go straight to 1.10 when it's out soon (1.10.0b1 is out)

[durack1]

@dnadeau4 just FYI numpy 1.10.0 just landed so should be available for starting to migrate cdms2/3 across..

[durack1]

@dnadeau4 - fixed by #1607?

[doutriaux1]

@durack1 eventually yes.

[dnadeau4]

@durack1 I am trying to read all referenced ticket which bring me to more referenced tickets. At this point I am not sure what the problem is with 1.10.0. Can we discuss this?

[durack1]

@dnadeau4 the issue that @doutriaux1 encountered with an earlier numpy release (1.9.2 from memory) was that the cdms2TransientVariable.mean() function wasn't returning a valid result.. The comment is documented in #915

[dnadeau4]

They are gaps in many data test due to numpy interpolation bug. This is why many "diags" testing are failing with this numpy version. I can't update to 1.10.1 until this is fixed. numpy/numpy#6468

[durack1]

@dnadeau4 this link (discussing cython vs numba and speedups using each) was interesting.. https://medium.com/@elgehelge/the-5-most-important-python-data-science-advancements-of-2015-a136482da89b#.4psi1vuqy

[durack1]

@dnadeau4 this was closed by CDAT/cdms#59, but the comment above link will be useful during the numba implementation.. If you have a numba issue already open it would be useful to copy this across