Skip to content

Instantly share code, notes, and snippets.

@bhawkins
Last active December 15, 2016 20:52
Show Gist options
  • Star 1 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save bhawkins/5095558 to your computer and use it in GitHub Desktop.
Save bhawkins/5095558 to your computer and use it in GitHub Desktop.
Check the performance of various ways of copying a numpy array.
#!/usr/bin/env python
from __future__ import print_function
import timeit
import numpy as np
from distutils.version import StrictVersion
from six.moves import range
setup = """
import numpy as np
from six.moves import range
n = 1000
x = np.random.random(n)
y = np.empty_like(x)
"""
# These methods modify y so that all its values are equal to those in x.
methods = [
"for i in range(n): y[i] = x[i]",
"y[:] = x"
]
if StrictVersion(np.__version__) >= StrictVersion('1.7.0'):
methods.append("np.copyto(y, x)")
# These methods create a new object y with values equal to those in x.
methods.extend(
"""
y = np.empty_like(x)
y[:] = x
#
y = np.empty(n)
y[:] = x
#
y = np.zeros_like(x)
y[:] = x
#
y = np.zeros(n)
y[:] = x
#
y = np.copy(x)
#
y = np.array(x)
#
y = 1*x
""".split('#\n')
)
niter = 10000
for method in methods:
print(method.strip())
print('Time =', timeit.timeit(method, setup=setup, number=niter))
print()
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment