A good article on optimization
http://scikit-learn.sourceforge.net/dev/developers/performance.html from our scikits-learn brethren. And here is a google summer of code 2007 participant: http://www.python.org/dev/peps/pep-0368 I'm interested in how much one would gain in comparison with a numpy based approach.
On Sat, Jun 4, 2011 at 1:45 PM, Pieter Holtzhausen <holtzhau@gmail.com>wrote:
http://scikit-learn.sourceforge.net/dev/developers/performance.html from our scikits-learn brethren.
Yes, that's a very good approach.
And here is a google summer of code 2007 participant: http://www.python.org/dev/peps/pep-0368 I'm interested in how much one would gain in comparison with a numpy based approach.
Eh, nothing? Looks like that never got off the ground, it sounds to me like it wanted to do what the buffer protocol does - just in a more clumsy way. I also don't see anything in that PEP that would make it faster then scipy.ndimage. Cheers, Ralf
On Sat, Jun 4, 2011 at 1:45 PM, Pieter Holtzhausen <holtzhau@gmail.com> wrote:
http://scikit-learn.sourceforge.net/dev/developers/performance.html from our scikits-learn brethren.
It's a pity this page doesn't yet document the valgrind / kcachegrind approach; I find this to be by far the most useful way of examining Cython bottlenecks! Graphical visualisation of Python code: $ easy_install RunSnakeRun $ python -m cProfile -o prof.data scriptname $ runsnake prof.data Graphical visualisation of extensions: $ valgrind --tool=callgrind -v --dump-instr=yes --trace-jump=yes --callgrind-out-file=callgrind.log python benchmark.py $ kcachegrind callgrind.log Cheers Stéfan
participants (3)
-
Pieter Holtzhausen -
Ralf Gommers -
Stéfan van der Walt