[SciPy-Dev] data rankings

Moritz Emanuel Beber moritz.beber@googlemail....
Sat Nov 19 06:36:55 CST 2011


Hi,

>
> rankdata is also pretty slow.  I have a cythonized version working 
> (along with tiecorrection), but it is not quite ready for a pull request.
rankdata currently uses a lot of look-ups, so far I don't see how they 
can be avoided. I used numpy.unique for one of the rankings so it 
doesn't need those look-ups but it's actually a bit slower so I guess 
they're done internally.

> If there is interest, I don't see any reason to not include more 
> ranking functions in scipy.  Moritz, what are the use-cases for the 
> ranking that you used?
>
I use the ranking to compare some dynamical data on different networks. 
The data are on different scales and cover varying orders of magnitude 
so I used a ranking to make them at all comparable. It's quite trivial 
really but I wanted ties to have the same rank no matter how many 
duplicates there are.

Best,
Moritz


More information about the SciPy-Dev mailing list