[Numpy-discussion] ufunc oddities
Stéfan van der Walt
stefan at sun.ac.za
Sun May 25 16:14:38 EDT 2008
2008/5/25 Anne Archibald <peridot.faceted at gmail.com>:
> 2008/5/25 Charles R Harris <charlesr.harris at gmail.com>:
>
>> So, please tell me how numpy is supposed to work. Write as much as you
>> please. If you are so moved, why not write the tests for all 64 ufuncs for
>> all types and combinations and verify that they are all correct as specified
>> and raise errors when they should.
>
> This sounds like something that will be easier with nose (if I
> understand nose correctly). Since the ufuncs are so uniform, it seems
> like the zillion tests you are proposing writing should be expressed
> as a list of ufuncs with some type annotations, a list of types to run
> them against, and a programmatic expression of the type rules Robert
> is proposing. While you can do this with the current test framework,
> you get one test rather than many, and the failure report is not
> nearly so valuable.
We do have ParametricTestCase, but yes -- nose makes this very easy.
Regards
Stéfan
More information about the NumPy-Discussion
mailing list