[SciPy-dev] Ball Tree code (ticket 1048)

Sturla Molden sturla@molden...
Tue Nov 17 05:45:41 CST 2009


Anne Archibald skrev:
 >
 > * The code is specific to Euclidean distances.

I was wondering if we should try to loop-unroll minkowski distances in 
cKDTree? Right now it does a lot of wasteful for work for common cases 
like Euclidian distances.

I think this version should be rather friendly to autovectorizing 
compilers. It's easy to change the hard-coded values of k and p by 
editing minkowski.py.

Also, absolute values are computed by zeroing the sign bit, which is 
much faster.

Sturla Molden
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: minkowski.py
Url: http://mail.scipy.org/pipermail/scipy-dev/attachments/20091117/11df4b32/attachment-0002.pl 
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: minkowski.pyx
Url: http://mail.scipy.org/pipermail/scipy-dev/attachments/20091117/11df4b32/attachment-0003.pl 


More information about the Scipy-dev mailing list