search_first_benchmark.py

"""
http://stackoverflow.com/questions/7632963/numpy-find-first-index-of-value-fast

First you need to run:
	f2py -c -m search search.f90
"""

from time import time
from matplotlib.pyplot import subplots, show
from numpy import array, nonzero, argwhere
import search


print(search.find_first.__doc__)

D = array(range(1000000))

def find_first(vec, needle):
	for k, val in enumerate(vec):
		if val == needle:
			return k
	return -1

needles = [1, 10, 100, 1000, 10000, 100000, 1000000-1, 2000000]
out = []

a = search.find_first(10, D)
reps = 100

for needle in needles:
	print('benchmark for needle at {0:d}'.format(needle))
	res = []

	t = time()
	for k in range(reps):
		try:
			a = argwhere(D==needle)[0][0]
		except IndexError:
			a = -1
	res.append(time() - t)

	t = time()
	for k in range(reps):
		try:
			a = nonzero(D == needle)[0][0]
		except IndexError:
			a = -1
	res.append(time() - t)

	t = time()
	for k in range(reps):
		a = (D == needle).tostring().find('\x01')
	res.append(time() - t)

	t = time()
	for k in range(reps):
		a = find_first(D, needle)
	res.append(time() - t)

	t = time()
	for k in range(reps):
		a = search.find_first(needle, D)
	res.append(time() - t)

	out.append(res)


outa = array(out) / reps
names = ['argwhere', 'nonzero', 'tostring', 'pyloop', 'f2py']

fig, ax = subplots(figsize=(5, 4))
fig.tight_layout()
for k, col in enumerate(outa.T):
	ax.loglog(needles, col, label=names[k])
ax.legend(loc='upper left')


if __name__ == '__main__':
	show()

Fortran code: https://gist.github.com/mverleg/71dbbf755ed1cf96589236a69bc64027