Chainerのexampleのmnistをcythonを使ってどれだけ高速化できるか検証してみた ref: http://qiita.com/GushiSnow/items/249d8a74c53dfb97c0ef

file0.txt

 python pycallgraph graphviz -- ./ファイル名.py

file1.txt

python -m cProfile

file2.txt

Profile.prof

         37494628 function calls (35068627 primitive calls) in 755.154 seconds

   Ordered by: internal time

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
    72000  448.089    0.006  448.651    0.006 adam.py:27(update_one_cpu)
   114000  187.057    0.002  187.057    0.002 {method 'dot' of 'numpy.ndarray' objects}
   216000   31.576    0.000   31.576    0.000 {method 'fill' of 'numpy.ndarray' objects}
    12000   23.122    0.002  163.601    0.014 variable.py:216(backward)

file3.txt

Profile.prof

         37466504 function calls (35040503 primitive calls) in 800.453 seconds

   Ordered by: internal time

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
    72000  473.638    0.007  474.181    0.007 adam.py:27(update_one_cpu)
   114000  199.589    0.002  199.589    0.002 {method 'dot' of 'numpy.ndarray' objects}
   216000   33.706    0.000   33.706    0.000 {method 'fill' of 'numpy.ndarray' objects}
    12000   24.754    0.002  173.816    0.014 variable.py:216(backward)
    28000    9.944    0.000   10.392    0.000 

file4.txt

/Users/smap2/.pyxbld/temp.macosx-10.10-x86_64-3.4/pyrex/train_mnist_c2.c:242:10: fatal error: 'numpy/arrayobject.h' file not found

file5.txt

/usr/local/include/

file6.txt

find / -name arrayobject.h -print 2> /dev/null

file7.txt

Profile.prof

         37466756 function calls (35040748 primitive calls) in 776.901 seconds

   Ordered by: internal time

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
    72000  458.284    0.006  458.812    0.006 adam.py:27(update_one_cpu)
   114000  194.834    0.002  194.834    0.002 {method 'dot' of 'numpy.ndarray' objects}
   216000   33.120    0.000   33.120    0.000 {method 'fill' of 'numpy.ndarray' objects}
    12000   24.025    0.002  168.772    0.014 variable.py:216(backward)

file8.txt

Profile.prof

         37250749 function calls (34824741 primitive calls) in 727.414 seconds

   Ordered by: internal time

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
    72000  430.495    0.006  430.537    0.006 optimizer.py:388(update_one)
   114000  180.775    0.002  180.775    0.002 {method 'dot' of 'numpy.ndarray' objects}
   216000   30.647    0.000   30.647    0.000 {method 'fill' of 'numpy.ndarray' objects}
    12000   21.766    0.002  157.230    0.013 variable.py:216(backward)