mmfutils.performance

• mmfutils.performance.blas
• mmfutils.performance.fft
  • fft()
  • fftfreq()
  • fftn()
  • get_fft()
  • get_fftn()
  • get_ifft()
  • get_ifftn()
  • ifft()
  • ifftn()
  • resample()
• mmfutils.performance.numexpr
• mmfutils.performance.threads
  • set_num_threads()

Tools for high-performance computing.

This module may rely on many other packages that are not easy to install such as pyfftw and the corresponding fftw implementation.

mmfutils.performance.auto_timeit(stmt='pass', setup='pass', time=1.0, max_time=30, repeat=5, globals=None, display=True, format='{min_t: 8.4f} {unit} per loop (best of {repeat}, {number} loops)')

Custom wrapper for the timeit module.

Parameters:

• time (float) – Desired execution time (s). Will be used to determine the number of times to run the code.

• max_time (float) – If running the code takes more than this amount of time, then only a single run will be used.

• display (bool) – If True, then print the results.

• format (str) – Format string for printing.

Return type:

(ts, number, factor, unit) if not display, else None.

Examples

>>> import time
>>> def f(t):
...     time.sleep(t)
>>> auto_timeit('f(t)', globals=dict(t=0.2, f=f))
20...ms per loop (best of 5, 1 loops)

You can increase the time if you don’t mind waiting, and get better accuracy.

>>> auto_timeit('f(t)', time=0.6, repeat=2, globals=dict(t=0.1, f=f))
1...ms per loop (best of 2, 2 loops)

If the test would exceed the maximum time, it simply uses a single loop:

>>> auto_timeit('f(t)', max_time=0.1, globals=dict(t=0.05, f=f))
5...ms per loop (best of 1, 1 loops)

You can also get the results programmatically to do your own analysis:

>>> ts, number, factor, unit = auto_timeit(
...     'f(t)', display=False, globals=dict(t=0.05, f=f))
>>> len(ts), float(min(ts)), float(np.median(ts)), factor, unit
(5, 0.05..., 0.05..., 1000, 'ms')