>> With the help of an awesome python community I have been able to pick up the
>> language and now willing to explore other cool extensions of it. > >Good work! > >> >> I routinely have large loops which could be ported to cython for speed. >> However I have never written a single line of cython code. Any pointers on >> getting started. > >There are two reasons for using cython: >1) To interface directly with existing C libraries. >2) To speed up CPU-intensive code. > >It sounds like you're interested in case 2). However, not all loops >benefit from cythonisation. Loosely cython is good when >a) you're doing something that can be written in a small amount of >efficient C code >b) but the corresponding Python code involves a lot of repeated >function calls or expression evaluations. > >If you're already familiar with C then you'll probably have some idea >when a) and b) apply. I would say that a prerequisite for learning to >speed up CPU-intensive code with cython would be learning to use the >python profilers. In particular you should learn to use cProfile and >timeit: (and of course pstats) I recently used Cython for the first time and I found it surprisingly easy. The installation under Linux is easy, but seems to be not really trivial under Windows (I never tried installing it under windows though; I'd first try one of the unofficial binaries: http://www.lfd.uci.edu/~gohlke/pythonlibs/#cython). One gotcha that I only found out later: it is possible to generate an html report of your Cythonized code. The yellower the code is coloured, the more it might still benefit from speed improvements: http://stackoverflow.com/questions/11058933/cython-a-flag-to-generate-yellow-shaded-html-without-command-line It's pretty cool to use cProfile to make a Cython-Python comparison of the function you're trying to speed up. In my case it was 87 % faster IIRC (this difference was not only significant, but also relevent: a use case where this would save 15 minutes is totally realistic) . Yaay! Albert-Jan _______________________________________________ Tutor maillist - Tutor@python.org To unsubscribe or change subscription options: http://mail.python.org/mailman/listinfo/tutor
