Check the performance of various ways of copying a numpy array.

time-numpy-deep-copy.py

#!/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()