Increase the speed of the Monte Carlo method of Cython cut-out implementation.

Today is Cython

I wrote before Rapidly implement the Monte Carlo method in Python.

The implementation using Cython was too omission, so I will supplement it.

This is the original Python code

import random
NUM=100000000
def monte():
    counter = 0
    for i in range(NUM):
        x = random.random()
        y = random.random()
        if x*x+y*y < 1.0:
            counter += 1
    pi = 4.0*counter/NUM
    print(pi)


def main():
    monte()

if __name__ == '__main__':
    main()

This is Cython

#monte.pyx
import random

cdef int NUM = 100000000

cdef cmonte():
    cdef :
        int counter = 0
        int i=0
        double x
        double y
    for i in range(NUM):
        x = random.random()
        y = random.random()
        if x*x + y*y < 1.0:
            counter += 1

    cdef double pi = 4.0*counter/NUM
    return pi

def monte():
    pi=cmonte()
    print(pi)

It's good that it shortened from about 100 seconds to 17 seconds (measured with a 12inch MacBook), but I should be able to do my best. Since the random part is Python code, it seems good to use the C implementation here.

Improved Cython

from libc.stdlib cimport rand, RAND_MAX

cdef int NUM = 100000000

def monte():
    cdef :
        int counter = 0
        int i=0
        double x
        double y
    for i in range(NUM):
        x = (rand()+1.0)/(RAND_MAX+2.0)
        y = (rand()+1.0)/(RAND_MAX+2.0)
        if x*x + y*y < 1.0:
            counter += 1

    pi = 4.0*counter/NUM
    print(pi)

Let's also write setup.py.

#setup.py
from setuptools import setup, Extension

ext_modules = [
    Extension(
        name='monte',
        sources=['monte.pyx']
        )
]

setup(
    name = 'cymonte',
    ext_modules = ext_modules
)

You can create a module by executing the following in the terminal. I think it can be done on Windows.

$ python setup.py build_ext --inplace

After that, write the main script to run it.

#main.py
import monte
monte.monte()

Let's move it.

$ time python main.py
real	0m2.081s
user	0m1.935s
sys 	0m0.070s

I tuned it and it's much faster! You were able to get close enough to the implementation of Numba and the implementation of C ++.

Reference:

Monte Carlo Simulation with Cython