Hi Serge. I'm just back in the UK from teaching High Performance Python to PhDs in Denmark last week. We covered Pythran (mainly Cython, also PyPy and Numba). For the above benchmarks what is the main reason for the speed improvements over numpy? Is it because you're using locally defined variants of the numpy functions? Is it because you have parallelisation (there's no note about openmp/multi-core)? Cheers, Ian. On 31 May 2014 16:46, serge Guelton <serge.guelton@xxxxxxxxxxxxxxxx> wrote: > Hi pythraners, > > Here is the result of a recent run of numpy-benchmarks. Enjoy! > > numba parakeet pypy python pythran > allpairs_distances 0.0 1.78 0.0 38.3 *1.71* > allpairs_distances_loops 57.7 1.78 0.0 50.3 *1.66* > arc_distance 1.37 1.6 0.0 *1.36* 1.84 > conv 2730.0 2.03 0.0 1900.0 *1.86* > create_grid 4.06 0.0 10.4 3.93 *3.85* > cronbach 1.65 0.0 0.0 1.64 *1.49* > diffusion 24.2 21.2 0.0 22.5 *3.99* > evolve 0.0 0.0 0.0 5.82 *3.53* > fdtd 0.0 0.0 0.0 1620.0 *1.85* > fft 0.0 0.0 0.0 24.9 *1.21* > grouping 0.0 0.0 0.0 2.1 *0.954* > growcut 7.49 *2.19* 0.0 1850.0 5.88 > harris 5.28 5.06 0.0 5.55 *2.79* > hasting 0.0119 0.0585 0.0 0.00884 *0.00129* > hyantes 287.0 *1.87* 0.0 247.0 2.46 > julia 223.0 2.94 0.0 2680.0 *2.59* > l2norm 5.79 12.4 0.0 6.15 *1.2* > local_maxima 0.0 0.0 0.0 58.6 *1.32* > mandel 4.5 0.0 17.7 452.0 *4.33* > multiple_sum 3.22 8.16 0.0 3.01 *1.13* > pairwise 8.9 3.7 0.0 3780.0 *3.58* > periodic_dist 2.08 0.0 0.0 1.78 *1.14* > repeating 1.21 0.0 0.0 1.26 *0.292* > reverse_cumsum 2.62 0.0 0.0 2.61 *2.36* > rosen 20.1 5.73 34.2 14.8 *1.39* > slowparts 0.0 1.23 2.6 5.81 *1.11* > smoothing 4.89 6.23 8.24 941.0 *4.83* > specialconvolve 0.0 0.0 0.0 *10.2* 0.0 > vibr_energy 2.45 2.45 0.0 2.57 *1.96* > wave 0.0 0.0 0.0 52.0 *1.23* > wdist 89.6 *1.45* 0.0 82.8 2.1 > > -- Ian Ozsvald (A.I. researcher) ian@xxxxxxxxxxxxxx http://IanOzsvald.com http://ModelInsight.io http://MorConsulting.com http://Annotate.IO http://SocialTiesApp.com http://TheScreencastingHandbook.com http://FivePoundApp.com http://twitter.com/IanOzsvald http://ShowMeDo.com