bmtools 0.2.1

System informations

bmtools provides some tools dedicated to benchmarking.

Requirements

python:

>= 3.7

matplotlib:

>= 3.0

numpy:

>= 1.1

Installation

Clone the github repo and

$ python setup.py install

or install via Pypi

$ pip install bmtools

Compare execution times

Benchmarking functions execution can be done with `Compare` class as follows:

import numpy as np
from bmtools import Compare

def star_op(x):
    """ Double star operator. """
    return x**0.5

def pow_op(x):
    """ pow function. """
    return pow(x, 0.5)

def sqrt_op(x):
    """ numpy.sqrt function. """
    return np.sqrt(x)

if __name__ == "__main__":

    # Single comparison
    bm1 = Compare(pow_op, star_op, sqrt_op)
    bm1.run_single(fargs=(np.random.rand(1000000), ))
    bm1.display()

    # Parametric comparison
    bm2 = Compare(pow_op, star_op, sqrt_op, unit='ms')
    for n in [2**n for n in range(16, 23)]:
        bm2.run_single(fargs=(np.random.rand(n), ), desc=n)

    bm2.display()
    bm2.bars()

+------------+---------------+----------------+----------------+-------+
| Function   |  Description  | Runtime [msec] |   Std [msec]   | Equal |
+------------+---------------+----------------+----------------+-------+
| pow_op     |      --       |    1.56256     |    0.00798     |  R1   |
| star_op    |      --       |    1.55787     |    0.00752     | ==R1  |
| sqrt_op    |      --       |    1.58628     |    0.04214     | ==R1  |
+------------+---------------+----------------+----------------+-------+

(...)

Compare provides three ways to display results:

• As a simple plot with the Compare.plot() method

• As a bar chart with the Compare.bar() method

• As a text table with the Compare.display() method

Compare also provides the parameters decorator to specify a list of args/kwarg that have to be passed to a function for parametric study. The the Compare.run_parametric method performs the comparison:

from bmtools import Compare

@Compare.parameters((1, 2,), (2, 3, ), x=(1, 10))
def op1(a, b, x=1):
    return a*x + b

@Compare.parameters((1, 2,), (2, 3,), x=(1, 10))
def op2(a, b, x=1):
    return a*x + b

if __name__ == "__main__":
    bm3 = Compare(op1, op2, unit='nsec')
    bm3.run_parametric()
    bm3.display()

+------------+---------------+----------------+----------------+-------+
| Function   |  Description  | Runtime [nsec] |   Std [nsec]   | Equal |
+------------+---------------+----------------+----------------+-------+
| op1        |   1, 2, x=1   |     359.8      |      12.0      |  R1   |
| op2        |   1, 2, x=1   |     354.5      |      8.4       | ==R1  |
+------------+---------------+----------------+----------------+-------+
| op1        |  1, 2, x=10   |     352.5      |      6.1       |  R2   |
| op2        |  1, 2, x=10   |     351.2      |      8.6       | ==R2  |
+------------+---------------+----------------+----------------+-------+

Time instance methods

The `mtimer` decorator can be used to time instance methods as follows:

import time
from bmtools import mtimer


class MtimeExample:
    """ mtimer examples. """

    def __init__(self):
        self.string = 'mtimer example'

    @mtimer(name='with arg')
    def method1(self, string):
        """ Example with argument. """
        time.sleep(0.2)
        print(self.string, string)
        time.sleep(0.2)

    @mtimer
    def method2(self, string):
        """ Example without argument. """
        time.sleep(0.1)
        print(self.string, string)
        time.sleep(0.1)


if __name__ == "__main__":

    mt = MtimeExample()

    for _ in range(2):
        mt.method1('with argument')

    mt.method2('without argument')

    format_mtimer(mt)

Add time probes to your code

The `TimeProbes` class provide a way to time blocks of code. Note that this class is largely inspired by Bench-it.

bm = TimeProbes()        # Create our probes
time.sleep(0.1)
bm('example')            # Create a probe named 'example'
time.sleep(0.2)
bm()                     # Create a probe without name

with bm as my_context:  # Use probe as context manager.
    time.sleep(0.8)      # my_context will be the name of the probe

bm.display()            # Display times measured at probe locations

+-------------------------------------------------------------------------------------------------------+
|                                              TimeProbes                                               |
+ ---------- + ------------------------ + ---------- + ---------------- + ---------------- + ---------- +
| Makers     |        File:line         |  Function  | Avg time [msec]  |  Runtime [msec]  |  Percent   |
+ ---------- + ------------------------ + ---------- + ---------------- + ---------------- + ---------- +
| example    | test_probes_simple.py:33 |     --     |    167.75452     |    167.75452     |  14334.3   |
| Probe 1    | test_probes_simple.py:35 |     --     |    201.12324     |    201.12324     |  17185.6   |
| my_context | test_probes_simple.py:37 |     --     |    800.91822     |    800.91822     |  68436.9   |
+ ---------- + ------------------------ + ---------- + ---------------- + ---------------- + ---------- +

References

The `TimeProbes` class is largely inpired by Bench-it:

https://pypi.org/project/bench-it/