New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
NaT comparison change breaks pandas test suite #7019
Comments
Thanks for testing that. Better now than later... |
Unfortunately, this change breaks several user facing APIs in pandas. We'll need to roll this change back for the NumPy v1.11 release, and preferably insert a deprecation warning in its place. |
but I agree with @shoyer you might consider providing a |
Yeah, that'd probably be more virtuous of us anyway... On Fri, Jan 15, 2016 at 1:15 PM, Jeff Reback notifications@github.com
Nathaniel J. Smith -- http://vorpus.org |
@jreback We are going to an accelerated release schedule. How long a time would you like to see a |
ideally only 1 major release cycle. what is your expected cycle time for major releases? we have not been good about that and have been leaving |
@shoyer Do you want to take care of this? Sounds like we need to revert the merge and put in place a FutureWarning. |
If not, let me know and I'll do it. |
Is there a cleaner way to handle this than putting the deprecate inside that ufunc kernel? That seems likely to have performance consequences. On Fri, Jan 15, 2016 at 5:27 PM, Charles Harris notifications@github.com
|
Oh, that's nasty. We don't want it in the loop if it can be helped, we would need to acquire the GIL. OTOH, we would need to check for NAT anyway if we want the warning to be specific. The only thing that occurs to me is to maybe make a wrapper function that shadows the true function, but that could be tricky. what with promotions and all and probably wouldn't make things any faster. Or we could just live with it for a release cycle. Hmm... If anyone has ideas don't hesitate to make a suggestion. |
We are shooting for 3-4 months. |
I guess we should just put it in the inner loop and pay the cost -- we'll On Fri, Jan 15, 2016 at 6:34 PM, Charles Harris notifications@github.com
Nathaniel J. Smith -- http://vorpus.org |
This is what I'm thinking at the moment
|
Sounds good to me
|
@charris Yes, I think that's what needs to be done. If you're able to do that quickly that would be appreciated -- if not I should be able to find some time to fix this tomorrow. |
I can do the reversion tonight, but I'm not going to try anything that needs the little gray cells. If you can't get to it tomorrow let me know and I'll give it a shot. The 1.11 branch may be a day or two late in any case. |
@shoyer How does your time look? I can do the |
Just for reference
|
See #7057. |
In Numpy 1.13 the plan is for NAT comparisons to behave like NaN comparisons, e.g., False except for 'NAT != NAT', which will be True. See the discussion at numpygh-7019 for details.
On track for 1.13, closing this. |
In Numpy 1.13 the plan is for NAT comparisons to behave like NaN comparisons, e.g., False except for 'NAT != NAT', which will be True. See the discussion at numpygh-7019 for details.
xref pandas-dev/pandas#12049
The source issue is changing NaT comparison (#7001)
I'll take a look and see if this is user facing or just in the test suite. If it does effect users, we may need to do a deprecation cycle instead.
The text was updated successfully, but these errors were encountered: