Measuring the per-item time for various numpy operations, after correcting for overhead

per-op-times.py

# Example output:
#   a + a: 1.181 ns/item
#   a / a: 2.577 ns/item
#   a ** a: 15.259 ns/item
#   np.log(a): 28.241 ns/item
#   np.sin(a): 22.202 ns/item
#   sp.gammaln(a): 40.876 ns/item
#   sp.erf(a): 21.297 ns/item

import timeit

LOOPS = 1000000

ARR_SIZES = [1, 1001]
SETUP_CODE = ("import numpy as np; "
              "import scipy.special as sp; "
              "a = np.asarray([1.] * %r)")
OPS = ["a + a",
       "a / a",
       "a ** a",
       "np.log(a)",
       "np.sin(a)",
       "sp.gammaln(a)",
       "sp.erf(a)",
       ]

for op in OPS:
    times = []
    for arr_size in ARR_SIZES:
        setup = SETUP_CODE % (arr_size,)
        times.append(timeit.timeit(op, setup=setup, number=LOOPS) / LOOPS)
    print("%s: %0.3f ns/item"
          % (op,
             (times[1] - times[0]) * 1.0
             / (ARR_SIZES[1] - ARR_SIZES[0]) * 1e9,
             ))