numpy-copy-mab.py

#!python
# Adapted from StackOverflow user mab
# http://stackoverflow.com/a/33672015/112699
import timeit
import numpy as np
import pandas as pd
from IPython.display import display

def profile_this(methods, setup='', n=100, niter=10**4):
    print('n: {0}, niter: {1:.0e}'.format(n, niter))
    timings = np.array([timeit.timeit(method, setup=setup, number=niter) for method in methods])
    ranking = np.argsort(timings)
    timings = np.array(timings)[ranking]
    methods = np.array(methods)[ranking]
    speedups = np.max(timings) / timings

    pd.set_option('html', False)
    data = {'time (s)': timings, 'speedup': ['{0:0.2f}x'.format(s) if 1 != s else '' for s in speedups], 'methods': methods}
    data_frame = pd.DataFrame(data, columns=['time (s)', 'speedup', 'methods'])

    display(data_frame)
    print()

setup_template = '''import numpy as np; n={n}; x = np.random.random(n)'''
methods = (
      '''y = np.zeros(n, dtype=x.dtype); y[:] = x''',
      '''y = np.zeros_like(x); y[:] = x''',
      '''y = np.empty(n, dtype=x.dtype); y[:] = x''',
      '''y = np.empty_like(x); y[:] = x''',
      '''y = np.copy(x)''',
      '''y = x.astype(x.dtype)''',
      '''y = 1*x''',
      '''y = np.empty_like(x); np.copyto(y, x)''',
      '''y = np.empty_like(x); np.copyto(y, x, casting='no')''',
      '''y = np.empty(n)\nfor i in range(x.size):\n\ty[i] = x[i]'''
)

for logn, logit in ((2, 6), (3, 6), (3.8, 6), (4, 6), (5, 5), (6, 4.5)):
    n, niter = int(10**logn), int(10**logit)
    setup = setup_template.format(n=n)
    if n > 100:
        profile_this(methods[:-1], setup, n, niter)
    else:
        profile_this(methods, setup, n, niter)

output.txt

n: 100, niter: 1e+06
    time (s) speedup                                            methods
0   0.377211  35.91x                              y = x.astype(x.dtype)
1   0.440208  30.77x                     y = np.empty_like(x); y[:] = x
2   0.449129  30.16x              y = np.empty_like(x); np.copyto(y, x)
3   0.647460  20.92x  y = np.empty_like(x); np.copyto(y, x, casting=...
4   0.722250  18.75x                                     y = np.copy(x)
5   0.836623  16.19x                                            y = 1*x
6   1.062869  12.74x           y = np.empty(n, dtype=x.dtype); y[:] = x
7   1.103768  12.27x           y = np.zeros(n, dtype=x.dtype); y[:] = x
8   2.268195   5.97x                     y = np.zeros_like(x); y[:] = x
9  13.545740          y = np.empty(n)\nfor i in range(x.size):\n\ty[...
()
n: 1000, niter: 1e+06
   time (s) speedup                                            methods
0  0.651657   4.44x                              y = x.astype(x.dtype)
1  0.749732   3.86x              y = np.empty_like(x); np.copyto(y, x)
2  0.773912   3.74x                     y = np.empty_like(x); y[:] = x
3  0.988279   2.93x  y = np.empty_like(x); np.copyto(y, x, casting=...
4  1.123788   2.58x                                     y = np.copy(x)
5  1.410723   2.05x                                            y = 1*x
6  1.475579   1.96x           y = np.empty(n, dtype=x.dtype); y[:] = x
7  1.508836   1.92x           y = np.zeros(n, dtype=x.dtype); y[:] = x
8  2.895328                             y = np.zeros_like(x); y[:] = x
()
n: 6309, niter: 1e+06
   time (s) speedup                                            methods
0  2.141714   3.00x                              y = x.astype(x.dtype)
1  2.203533   2.91x                     y = np.empty_like(x); y[:] = x
2  2.231979   2.88x              y = np.empty_like(x); np.copyto(y, x)
3  2.502381   2.57x                                     y = np.copy(x)
4  2.665101   2.41x  y = np.empty_like(x); np.copyto(y, x, casting=...
5  2.822849   2.27x           y = np.empty(n, dtype=x.dtype); y[:] = x
6  2.892670   2.22x                                            y = 1*x
7  4.391657   1.46x           y = np.zeros(n, dtype=x.dtype); y[:] = x
8  6.421201                             y = np.zeros_like(x); y[:] = x
()
n: 10000, niter: 1e+06
   time (s) speedup                                            methods
0  3.263873   2.81x                              y = x.astype(x.dtype)
1  3.274130   2.80x              y = np.empty_like(x); np.copyto(y, x)
2  3.412645   2.68x  y = np.empty_like(x); np.copyto(y, x, casting=...
3  3.607340   2.54x                     y = np.empty_like(x); y[:] = x
4  3.851676   2.38x                                     y = np.copy(x)
5  4.292577   2.13x           y = np.empty(n, dtype=x.dtype); y[:] = x
6  4.457260   2.06x                                            y = 1*x
7  6.469592   1.42x           y = np.zeros(n, dtype=x.dtype); y[:] = x
8  9.160425                             y = np.zeros_like(x); y[:] = x
()
n: 100000, niter: 1e+05
   time (s) speedup                                            methods
0  3.588155   2.25x                     y = np.empty_like(x); y[:] = x
1  3.637090   2.22x  y = np.empty_like(x); np.copyto(y, x, casting=...
2  3.673425   2.19x                                     y = np.copy(x)
3  3.706141   2.18x                              y = x.astype(x.dtype)
4  3.729351   2.16x           y = np.empty(n, dtype=x.dtype); y[:] = x
5  3.785925   2.13x              y = np.empty_like(x); np.copyto(y, x)
6  4.758041   1.69x                                            y = 1*x
7  6.385678   1.26x           y = np.zeros(n, dtype=x.dtype); y[:] = x
8  8.061876                             y = np.zeros_like(x); y[:] = x
()
n: 1000000, niter: 3e+04
    time (s) speedup                                            methods
0  17.355430   1.95x  y = np.empty_like(x); np.copyto(y, x, casting=...
1  17.450905   1.94x                              y = x.astype(x.dtype)
2  17.702589   1.91x                     y = np.empty_like(x); y[:] = x
3  17.876651   1.89x              y = np.empty_like(x); np.copyto(y, x)
4  18.069141   1.87x           y = np.empty(n, dtype=x.dtype); y[:] = x
5  18.179504   1.86x                                     y = np.copy(x)
6  22.717311   1.49x                                            y = 1*x
7  29.783388   1.14x           y = np.zeros(n, dtype=x.dtype); y[:] = x
8  33.870803                             y = np.zeros_like(x); y[:] = x
()
[Finished in 338.375s]