[LNT] r207898 - Use Mann-Whitney U test to identify changes

[Yi Kong]

In the docstring, it says "Use only when the n in each condition is < 20 
and you have 2 independent samples of ranks."

On 04/05/14 16:03, Anton Korobeynikov wrote:
> I looked over the sources and it looks like it only implements the
> asymptotic version via a normal approximation which may be used only
> when sample size is rather big (say > 20).
>
> So, this does not seem to be a proper alternative. And given that the
> webpage says that SciPy used the exactly same sources to start with,
> it may happen,
> that SciPy implementation is broken for small sample sizes as well...
>
> On Sun, May 4, 2014 at 6:50 PM, Yi Kong <Yi.Kong at arm.com> wrote:
>> There is a library that provides Mann-Whitney U test which is written purely
>> in Python and it's released in MIT-like license. Can we use that as a
>> replacement to SciPy?(I'm not familiar with licensing issues with LLVM)
>>
>> http://www.nmr.mgh.harvard.edu/Neural_Systems_Group/gary/python.html
>>
>>
>> On 03/05/14 22:40, Chris Matthews wrote:
>>>
>>> This package index will need to be updated before the bots will work:
>>>
>>> http://lab.llvm.org/packages/
>>>
>>> Their pip installations are limited to packages in that repo.
>>>
>>> On May 3, 2014, at 2:34 PM, Renato Golin <renato.golin at linaro.org> wrote:
>>>
>>>> On 3 May 2014 22:18, Chris Matthews <chris.matthews at apple.com> wrote:
>>>>>
>>>>> I would assume on the linux bots installing those packages + pip
>>>>> installing the scipy into the virtualenv will work?
>>>>
>>>>
>>>> Nope. I did install scipy (which pulled numpy), but that doesn't fix
>>>> the virtualenv.
>>>>
>>>> --renato
>>>
>>>
>>>
>>>
>>
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
>
>