An easy and convenient way to performance test blocks of python code. Tired of writing separate scripts for your performance tests? Don't like coding in strings? Using the pyperform decorators, you can easily implement timeit tests to your functions with just one line!
Features of pyperform include:
- Quick, easy to implement in-code performance tests that run once when the function is defined
- Speed comparison of several functions.
- Validation of results between ComparisonBenchmarks
- Summary reports.
- Supports class functions as well as global functions.
- Performance tests can easily be disabled/enabled globally.
- Community-driven library of performance tests to learn from.
To install:
pip install pyperform
PyPerform was developed in Python 2.7 but has been tested with Python 3.4. Please report any compatibility issues or send pull requests with your changes!
To use pyperform to benchmark functions, you need to add one of the following decorators:
@BenchmarkedFunction(setup=None,
classname=None,
largs=None,
kwargs=None,
timeit_repeat=3,
timeit_number=1000)
@BenchmarkedClass(setup=None,
largs=None,
kwargs=None,
timeit_repeat=3,
timeit_number=1000)
@ComparisonBenchmark(group,
classname=None,
setup=None,
largs=None,
kwargs=None,
validation=False,
timeit_repeat=3,
timeit_number=1000)
where largs is a list of arguments to pass to the function and kwargs is a dictionary of keyword arguments to pass to the function. The setup argument is described in the following section. All decorators have timeit_repeat and timeit_number arguments which are can be used to set the number of trials and repetitions to use with timeit. The ComparisonBenchmark has a validation flag, which when set to True, will attempt to compare the results of the functions in the group.
Sometimes your decorated function will require some setup code or imported modules. You can easily include any lines of
code by by appending the tag #!
to the end of the line. For functions and classes, you only need to tag the def
or
class
line and PyPerform will include the entire function/class definition as setup code.
For example:
from pyperform import BenchmarkedFunction
import math #!
a = 10 #!
def do_calcuation(a, b): #!
return a * b
@BenchmarkedFunction(largs=(5,))
def call_function(b):
# We can reference the `a` variable because it is tagged
result = a * b
assert result == 50
# We can call the math module because it is tagged.
math.log10(100)
# We can call this function because it is tagged.
calc_result = do_calcuation(a, b)
return calc_result
Results in:
call_function 6.214 us
All decorators have a setup argument which can be either a function with no arguments, or string of code. If given a function, the body of the function is executed in the global scope. This means that objects and variables instantiated in the body of the function are accessible from within the benchmarked function.
Example:
from pyperform import BenchmarkedFunction
def _setup():
a = 10
@BenchmarkedFunction(setup=_setup, largs=(5,))
def multiply_by_a(b):
result = a * b
assert result == 50
return result
Results in:
multiply_by_a 3.445 us
Pyperform will also work on class methods, but in order to do so, we must instantiate an instance of the class.
This is done in BenchmarkedClass
. Then once we have decorated the class with BenchmarkedClass
, we can use
ComparisonBenchmark
or BenchmarkedFunction
to performance test the class's methods.
Note that when benchmarking class methods, the classname
argument to ComparisonBenchmark must be provided.
This argument will hopefully be removed in the future.
In the BenchmarkedClass we instantiate a Person object and then run three benchmarked class-methods.
Two of the class-methods are ComparisonBenchmarks
and will be compared with one another. To see the result, you must
call the ComparisonBenchmark.summarize()
function. The third function is a duplicate of calculate_savings_method2 but
it is a BenchmarkedFunction instead. The result of BenchmarkedFunctions is printed when the script is run.
from pyperform import BenchmarkedClass, ComparisonBenchmark, BenchmarkedFunction
@BenchmarkedClass(largs=('Calvin', 24, 1000.,), kwargs={'height': '165 cm'})
class Person(object):
def __init__(self, name, age, monthly_income, height=None, *args, **kwargs):
self.name = name
self.age = age
self.height = height
self.monthly_income = monthly_income
@ComparisonBenchmark('Calculate Savings', classname="Person", timeit_number=100,
validation=True, largs=(55,), kwargs={'monthly_spending': 500})
def calculate_savings_method1(self, retirement_age, monthly_spending=0):
savings = 0
for y in range(self.age, retirement_age):
for m in range(12):
savings += self.monthly_income - monthly_spending
return savings
@ComparisonBenchmark('Calculate Savings', classname="Person", timeit_number=100,
validation=True, largs=(55,), kwargs={'monthly_spending': 500})
def calculate_savings_method2(self, retirement_age, monthly_spending=0):
yearly_income = 12 * (self.monthly_income - monthly_spending)
n_years = retirement_age - self.age
if n_years > 0:
return yearly_income * n_years
@BenchmarkedFunction(classname="Person", timeit_number=100,
largs=(55,), kwargs={'monthly_spending': 500})
def same_as_method_2(self, retirement_age, monthly_spending=0):
yearly_income = 12 * (self.monthly_income - monthly_spending)
n_years = retirement_age - self.age
if n_years > 0:
return yearly_income * n_years
You can print the summary to file or if ComparisonBenchmark.summarize() is not given an fs parameter, it will print to console.
report_file = open('report.txt', 'w')
ComparisonBenchmark.summarize(group='Calculate Savings', fs=report_file)
This results in a file report.txt
that contains the ComparisonBenchmark's results:
Call statement:
instance.calculate_savings_method2(55, monthly_spending=500)
Rank Function Name Time % of Fastest timeit_repeat timeit_number
------------------------------------------------------------------------------------------------------------------------
1 Person.calculate_savings_method2 267.093 ns 100.0 3 100
2 Person.calculate_savings_method1 35.623 us 0.7 3 100
------------------------------------------------------------------------------------------------------------------------
Source Code:
------------------------------------------------------------------------------------------------------------------------
def calculate_savings_method2(self, retirement_age, monthly_spending=0):
yearly_income = 12 * (self.monthly_income - monthly_spending)
n_years = retirement_age - self.age
if n_years > 0:
return yearly_income * n_years
------------------------------------------------------------------------------------------------------------------------
def calculate_savings_method1(self, retirement_age, monthly_spending=0):
savings = 0
for y in range(self.age, retirement_age):
for m in range(12):
savings += self.monthly_income - monthly_spending
return savings
------------------------------------------------------------------------------------------------------------------------
and printed to the screen, the results of the BenchmarkedFunction
same_as_method_2 262.827 ns
ComparisonBenchmark has a optional argument validate
. When validate=True
, the return value of each
ComparisonBenchmark in a group is compared. If the results of the function are the not same, a ValidationError is raised.
Example:
from pyperform import ComparisonBenchmark
from math import sin #!
@ComparisonBenchmark('Group1', validation=True, largs=(100,))
def list_append(n, *args, **kwargs):
l = []
for i in xrange(1, n):
l.append(sin(i))
return l
@ComparisonBenchmark('Group1', validation=True, largs=(100,))
def list_comprehension(n, *args, **kwargs):
return 1
Output:
pyperform.ValidationError: Results of functions list_append and list_comprehension are not equivalent.
list_append: [0.8414709848078965, 0.9092974268256817, 0.1411200080598672, -0.7568024953079282, -0.9589242746631385, -0.27941549819892586, 0.6569865987187891, 0.9893582466233818, 0.4121184852417566, -0.5440211108893698, -0.9999902065507035, -0.5365729180004349, 0.4201670368266409, 0.9906073556948704, 0.6502878401571168, -0.2879033166650653, -0.9613974918795568, -0.750987246771676, 0.14987720966295234, 0.9129452507276277, 0.8366556385360561, -0.008851309290403876, -0.8462204041751706, -0.9055783620066239, -0.13235175009777303, 0.7625584504796027, 0.956375928404503, 0.27090578830786904, -0.6636338842129675, -0.9880316240928618, -0.404037645323065, 0.5514266812416906, 0.9999118601072672, 0.5290826861200238, -0.428182669496151, -0.9917788534431158, -0.6435381333569995, 0.2963685787093853, 0.9637953862840878, 0.7451131604793488, -0.158622668804709, -0.9165215479156338, -0.8317747426285983, 0.017701925105413577, 0.8509035245341184, 0.9017883476488092, 0.123573122745224, -0.7682546613236668, -0.9537526527594719, -0.26237485370392877, 0.6702291758433747, 0.9866275920404853, 0.39592515018183416, -0.5587890488516163, -0.9997551733586199, -0.5215510020869119, 0.43616475524782494, 0.9928726480845371, 0.6367380071391379, -0.3048106211022167, -0.9661177700083929, -0.7391806966492228, 0.16735570030280691, 0.9200260381967906, 0.8268286794901034, -0.026551154023966794, -0.8555199789753223, -0.8979276806892913, -0.11478481378318722, 0.7738906815578891, 0.9510546532543747, 0.25382336276203626, -0.6767719568873076, -0.9851462604682474, -0.38778163540943045, 0.5661076368981803, 0.9995201585807313, 0.5139784559875352, -0.4441126687075084, -0.9938886539233752, -0.6298879942744539, 0.31322878243308516, 0.9683644611001854, 0.7331903200732922, -0.1760756199485871, -0.9234584470040598, -0.8218178366308225, 0.03539830273366068, 0.8600694058124533, 0.8939966636005579, 0.10598751175115685, -0.7794660696158047, -0.9482821412699473, -0.24525198546765434, 0.683261714736121, 0.9835877454343449, 0.3796077390275217, -0.5733818719904229, -0.9992068341863537]
list_comprehension: 1