Posts about Performance

RSS feed

Time the performance by timeit and cell magic

Comments

Goal

This post aims to introduce how to measure the running time of your code or function by using timeit module and cell magic command %%time and %%timeit.

Reference *Time a Python Function

Libraries

In [1]:
import pandas as pd
import timeit

Measure a literal code

In [2]:
timeit.timeit('2**10')
Out[2]:
0.01853936500265263

Measure the execution time for a function

In [4]:
def execute_abc(x=10):
    return 2 ** 10
In [7]:
timeit.timeit(execute_abc)
Out[7]:
0.0984989230055362

Measure the time by using Cell Magic in Jupyter Notebook

In [3]:
%%time
2**10

CPU times: user 4 µs, sys: 1e+03 ns, total: 5 µs
Wall time: 11 µs
Out[3]:
1024
In [9]:
%timeit 2**10

17.9 ns ± 0.597 ns per loop (mean ± std. dev. of 7 runs, 100000000 loops each)

Performance difference between append and insert in Python

Comments

Goal

This post aims to compare the performance between append and insert in Python. The performance comparison is simply done by the piece of code that counts a number, append it to a list, and then reverse it.

We will see the significant difference between two codes: one using append is linear and another using insert is quadratic run time growth as below.

Reference

Libraries

In [26]:
from timeit import Timer
import pandas as pd
%matplotlib inline

Append

In [6]:
count = 10**5
In [9]:
def count_by_append(count):
    nums = []
    for i in range(count):
        nums.append(i)
    nums.reverse()
count_by_append(count)

The execution time is 22ms

Insert

In [10]:
def count_by_insert(count):
    nums = []
    for i in range(count):
        nums.insert(0, i)
count_by_insert(count)

The execution time is 3.53s

Comparison

In [24]:
counts = [10 ** i for i in range(5)]
time_by_append = []
time_by_insert = []

for count in counts:
    print(f'Processing {count}')
    t = Timer(lambda: count_by_append(count))
    time_by_append.append(t.timeit(number=10))
    t = Timer(lambda: count_by_insert(count))
    time_by_insert.append(t.timeit(number=10))

df_performance = pd.DataFrame({'count': counts,
                               'count_by_append': time_by_append,
                               'count_by_insert': time_by_insert})
df_performance

Processing 1
Processing 10
Processing 100
Processing 1000
Processing 10000
Out[24]:
count count_by_append count_by_insert
0 1 0.000013 0.000011
1 10 0.000019 0.000027
2 100 0.000107 0.000227
3 1000 0.000927 0.005318
4 10000 0.009438 0.379457
In [34]:
# Plot the performance difference
df_performance.set_index('count').plot(title='Performance Comparison beteen Append and Insert');