Use just like the magic %timeit in Ipython
>from itimeit import itimeit
>itimeit("1+1")
10000000 loops, best of 3: 21.03 ns per loop
Can also specify a baseline to subtract. For example,
specifying baseline="pass" let's you apprximately compensate
for the time taken by the timing loop itself.
>itimeit("1+1", baseline="pass")
10000000 loops, best of 3: 4.367 ns per loop
Access locals, including functions and imported modules, directly in the code to benchmark, without a cumbersome setup step.
>def foo(i):
while i != 0:
i /= 2
>itimeit("foo(100)")
1000000 loops, best of 3: 593.7 ns per loop
>itimeit("np.random.standard_normal((1000,1000))")
10 loops, best of 3: 72.01 ms per loop
The cost of this is that the environment is affected when timing mutating commands.
>a = np.arange(10)
>def bar(a): a += 1
1000000 loops, best of 3: 1.145 µs per loop
>print a[0]
3111111