6.13. Testing¶
6.13.1. pytest benchmark: A Pytest Fixture to Benchmark Your Code¶
!pip install pytest-benchmark
If you want to benchmark your code while testing with pytest, try pytest-benchmark.
To use pytest-benchmark works, add benchmark to the test function that you want to benchmark.
# pytest_benchmark_example.py
def list_comprehension(len_list=5):
return [i for i in range(len_list)]
def test_concat(benchmark):
res = benchmark(list_comprehension)
assert res == [0, 1, 2, 3, 4]
On your terminal, type:
$ pytest pytest_benchmark_example.py
Now you should see the statistics of the time it takes to execute the test functions on your terminal:
============================= test session starts ==============================
platform linux -- Python 3.8.10, pytest-6.2.5, py-1.10.0, pluggy-0.13.1
benchmark: 3.4.1 (defaults: timer=time.perf_counter disable_gc=False min_rounds=5 min_time=0.000005 max_time=1.0 calibration_precision=10 warmup=False warmup_iterations=100000)
rootdir: /home/khuyen/book/book/Chapter4
plugins: hydra-core-1.1.1, Faker-8.12.1, benchmark-3.4.1, repeat-0.9.1, anyio-3.3.0
collected 1 item
pytest_benchmark_example.py . [100%]
----------------------------------------------------- benchmark: 1 tests ----------------------------------------------------
Name (time in ns) Min Max Mean StdDev Median IQR Outliers OPS (Mops/s) Rounds Iterations
-----------------------------------------------------------------------------------------------------------------------------
test_concat 286.4501 4,745.5498 309.3872 106.6583 297.5001 5.3500 2686;5843 3.2322 162101 20
-----------------------------------------------------------------------------------------------------------------------------
Legend:
Outliers: 1 Standard Deviation from Mean; 1.5 IQR (InterQuartile Range) from 1st Quartile and 3rd Quartile.
OPS: Operations Per Second, computed as 1 / Mean
============================== 1 passed in 2.47s ===============================
6.13.2. pytest.mark.parametrize: Test Your Functions with Multiple Inputs¶
!pip install pytest
If you want to test your function with different examples, use pytest.mark.parametrize decorator.
To use pytest.mark.parametrize, add @pytest.mark.parametrize to the test function that you want to experiment with.
# pytest_parametrize.py
import pytest
def text_contain_word(word: str, text: str):
'''Find whether the text contains a particular word'''
return word in text
test = [
('There is a duck in this text',True),
('There is nothing here', False)
]
@pytest.mark.parametrize('sample, expected', test)
def test_text_contain_word(sample, expected):
word = 'duck'
assert text_contain_word(word, sample) == expected
In the code above, I expect the first sentence to contain the word “duck” and expect the second sentence not to contain that word. Let’s see if my expectations are correct by running:
$ pytest pytest_parametrize.py
============================= test session starts ==============================
platform linux -- Python 3.8.10, pytest-6.2.5, py-1.10.0, pluggy-1.0.0
benchmark: 3.4.1 (defaults: timer=time.perf_counter disable_gc=False min_rounds=5 min_time=0.000005 max_time=1.0 calibration_precision=10 warmup=False warmup_iterations=100000)
rootdir: /home/khuyen/book/book/Chapter4
plugins: benchmark-3.4.1, anyio-3.3.0
collecting ...
collected 2 items
pytest_parametrize.py .. [100%]
============================== 2 passed in 0.01s ===============================
Sweet! 2 tests passed when running pytest.
6.13.3. pytest parametrize twice: Test All Possible Combinations of Two Sets of Parameters¶
!pip install pytest
If you want to test the combinations of two sets of parameters, writing all possible combinations can be time-consuming and is difficult to read.
import pytest
def average(n1, n2):
return (n1 + n2) / 2
def perc_difference(n1, n2):
return (n2 - n1)/n1 * 100
# Test the combinations of operations and inputs
@pytest.mark.parametrize("operation, n1, n2", [(average, 1, 2), (average, 2, 3), (perc_difference, 1, 2), (perc_difference, 2, 3)])
def test_is_float(operation, n1, n2):
assert isinstance(operation(n1, n2), float)
You can save your time by using pytest.mark.parametrize twice instead.
# pytest_combination.py
import pytest
def average(n1, n2):
return (n1 + n2) / 2
def perc_difference(n1, n2):
return (n2 - n1)/n1 * 100
# Test the combinations of operations and inputs
@pytest.mark.parametrize("operation", [average, perc_difference])
@pytest.mark.parametrize("n1, n2", [(1, 2), (2, 3)])
def test_is_float(operation, n1, n2):
assert isinstance(operation(n1, n2), float)
On your terminal, run:
$ pytest -v pytest_combination.py
============================= test session starts ==============================
platform linux -- Python 3.8.10, pytest-6.2.5, py-1.10.0, pluggy-0.13.1 -- /home/khuyen/book/venv/bin/python3
cachedir: .pytest_cache
benchmark: 3.4.1 (defaults: timer=time.perf_counter disable_gc=False min_rounds=5 min_time=0.000005 max_time=1.0 calibration_precision=10 warmup=False warmup_iterations=100000)
hypothesis profile 'default' -> database=DirectoryBasedExampleDatabase('/home/khuyen/book/book/Chapter5/.hypothesis/examples')
rootdir: /home/khuyen/book/book/Chapter5
plugins: hydra-core-1.1.1, Faker-8.12.1, benchmark-3.4.1, repeat-0.9.1, anyio-3.3.0, hypothesis-6.31.6, typeguard-2.13.3
collected 4 items
pytest_combination.py::test_is_float[1-2-average] PASSED [ 25%]
pytest_combination.py::test_is_float[1-2-perc_difference] PASSED [ 50%]
pytest_combination.py::test_is_float[2-3-average] PASSED [ 75%]
pytest_combination.py::test_is_float[2-3-perc_difference] PASSED [100%]
============================== 4 passed in 0.27s ===============================
From the output above, we can see that all possible combinations of the given operations and inputs are tested.
6.13.4. Assign IDs to Test Cases¶
When using pytest parametrize, it can be difficult to understand the role of each test case.
# pytest_without_ids.py
from pytest import mark
def average(n1, n2):
return (n1 + n2) / 2
@mark.parametrize(
"n1, n2",
[(-1, -2), (2, 3), (0, 0)],
)
def test_is_float(n1, n2):
assert isinstance(average(n1, n2), float)
$ pytest -v pytest_without_ids.py
============================= test session starts ==============================
platform linux -- Python 3.8.10, pytest-6.2.5, py-1.10.0, pluggy-0.13.1 -- /home/khuyen/book/venv/bin/python3
cachedir: .pytest_cache
benchmark: 3.4.1 (defaults: timer=time.perf_counter disable_gc=False min_rounds=5 min_time=0.000005 max_time=1.0 calibration_precision=10 warmup=False warmup_iterations=100000)
hypothesis profile 'default' -> database=DirectoryBasedExampleDatabase('/home/khuyen/book/book/Chapter5/.hypothesis/examples')
rootdir: /home/khuyen/book/book/Chapter5
plugins: hydra-core-1.1.1, Faker-8.12.1, benchmark-3.4.1, repeat-0.9.1, anyio-3.3.0, hypothesis-6.31.6, cases-3.6.10, typeguard-2.13.3
collected 3 items
pytest_without_ids.py::test_is_float[-1--2] PASSED [ 33%]
pytest_without_ids.py::test_is_float[2-3] PASSED [ 66%]
pytest_without_ids.py::test_is_float[0-0] PASSED [100%]
============================== 3 passed in 0.26s ===============================
You can add ids to pytest parametrize to assign a name to each test case.
# pytest_ids.py
from pytest import mark
def average(n1, n2):
return (n1 + n2) / 2
@mark.parametrize(
"n1, n2",
[(-1, -2), (2, 3), (0, 0)],
ids=["neg and neg", "pos and pos", "zero and zero"],
)
def test_is_float(n1, n2):
assert isinstance(average(n1, n2), float)
$ pytest -v pytest_ids.py
============================= test session starts ==============================
platform linux -- Python 3.8.10, pytest-6.2.5, py-1.10.0, pluggy-0.13.1 -- /home/khuyen/book/venv/bin/python3
cachedir: .pytest_cache
benchmark: 3.4.1 (defaults: timer=time.perf_counter disable_gc=False min_rounds=5 min_time=0.000005 max_time=1.0 calibration_precision=10 warmup=False warmup_iterations=100000)
hypothesis profile 'default' -> database=DirectoryBasedExampleDatabase('/home/khuyen/book/book/Chapter5/.hypothesis/examples')
rootdir: /home/khuyen/book/book/Chapter5
plugins: hydra-core-1.1.1, Faker-8.12.1, benchmark-3.4.1, repeat-0.9.1, anyio-3.3.0, hypothesis-6.31.6, cases-3.6.10, typeguard-2.13.3
collected 3 items
pytest_ids.py::test_is_float[neg and neg] PASSED [ 33%]
pytest_ids.py::test_is_float[pos and pos] PASSED [ 66%]
pytest_ids.py::test_is_float[zero and zero] PASSED [100%]
============================== 3 passed in 0.27s ===============================
We can see that instead of [-1--2], the first test case is shown as neg and neg. This makes it easier for others to understand the roles of your test cases.
6.13.5. Pytest Fixtures: Use The Same Data for Different Tests¶
!pip install pytest
If you want to use the same data to test different functions, use pytest fixtures.
To use pytest fixtures, add the decorator @pytest.fixture to the function that creates the data you want to reuse.
# pytest_fixture.py
import pytest
from textblob import TextBlob
def extract_sentiment(text: str):
"""Extract sentimetn using textblob. Polarity is within range [-1, 1]"""
text = TextBlob(text)
return text.sentiment.polarity
@pytest.fixture
def example_data():
return 'Today I found a duck and I am happy'
def test_extract_sentiment(example_data):
sentiment = extract_sentiment(example_data)
assert sentiment > 0
On your terminal, type:
$ pytest pytest_fixture.py
Output:
============================= test session starts ==============================
platform linux -- Python 3.8.10, pytest-6.2.5, py-1.10.0, pluggy-1.0.0
benchmark: 3.4.1 (defaults: timer=time.perf_counter disable_gc=False min_rounds=5 min_time=0.000005 max_time=1.0 calibration_precision=10 warmup=False warmup_iterations=100000)
rootdir: /home/khuyen/book/book/Chapter4
plugins: benchmark-3.4.1, anyio-3.3.0
collected 1 item
pytest_fixture.py . [100%]
============================== 1 passed in 0.53s ===============================
6.13.6. Pytest skipif: Skip a Test When a Condition is Not Met¶
If you want to skip a test when a condition is not met, use pytest skipif. For example, in the code below, I use skipif to skip a test if the python version is less than 3.9.
# pytest_skip.py
import sys
import pytest
def add_two(num: int):
return num + 2
@pytest.mark.skipif(sys.version_info < (3, 9), reason="Eequires Python 3.9 or higher")
def test_add_two():
assert add_two(3) == 5
On your terminal, type:
$ pytest pytest_skip.py -v
Output:
============================= test session starts ==============================
platform darwin -- Python 3.8.10, pytest-7.1.2, pluggy-1.0.0 -- /Users/khuyen/book/venv/bin/python3
cachedir: .pytest_cache
rootdir: /Users/khuyen/book/Efficient_Python_tricks_and_tools_for_data_scientists/Chapter5
collecting ...
collected 1 item
pytest_skip.py::test_add_two SKIPPED (Eequires Python 3.9 or higher) [100%]
============================== 1 skipped in 0.01s ==============================
6.13.7. Pytest repeat¶
!pip install pytest-repeat
It is a good practice to test your functions to make sure they work as expected, but sometimes you need to test 100 times until you found the rare cases when the test fails. That is when pytest-repeat comes in handy.
To use pytest-repeat, add the decorator @pytest.mark.repeat(N) to the test function you want to repeat N times
# pytest_repeat_example.py
import pytest
import random
def generate_numbers():
return random.randint(1, 100)
@pytest.mark.repeat(100)
def test_generate_numbers():
assert generate_numbers() > 1 and generate_numbers() < 100
# pytest_repeat_example.py
import pytest
import random
def generate_numbers():
return random.randint(1, 100)
@pytest.mark.repeat(100)
def test_generate_numbers():
assert generate_numbers() > 1 and generate_numbers() < 100
On your terminal, type:
pytest pytest_repeat_example.py
We can see that 100 experiments are executed and passed:
============================= test session starts ==============================
platform linux -- Python 3.8.10, pytest-6.2.5, py-1.10.0, pluggy-1.0.0
benchmark: 3.4.1 (defaults: timer=time.perf_counter disable_gc=False min_rounds=5 min_time=0.000005 max_time=1.0 calibration_precision=10 warmup=False warmup_iterations=100000)
rootdir: /home/khuyen/book/book/Chapter4
plugins: benchmark-3.4.1, repeat-0.9.1, anyio-3.3.0
collected 100 items
pytest_repeat_example.py ............................................... [ 47%]
..................................................... [100%]
============================= 100 passed in 0.07s ==============================
6.13.8. pytest-sugar: Show the Failures and Errors Instantly With a Progress Bar¶
!pip install pytest-sugar
It can be frustrating to wait for a lot of tests to run before knowing the status of the tests. If you want to see the failures and errors instantly with a progress bar, use pytest-sugar.
pytest-sugar is a plugin for pytest. The code below shows how the outputs will look like when running pytest.
$ pytest
Test session starts (platform: linux, Python 3.8.10, pytest 6.2.5, pytest-sugar 0.9.4)
benchmark: 3.4.1 (defaults: timer=time.perf_counter disable_gc=False min_rounds=5 min_time=0.000005 max_time=1.0 calibration_precision=10 warmup=False warmup_iterations=100000)
rootdir: /home/khuyen/book/book/Chapter5
plugins: hydra-core-1.1.1, Faker-8.12.1, benchmark-3.4.1, repeat-0.9.1, anyio-3.3.0, sugar-0.9.4
collecting ...
pytest_sugar_example/test_benchmark_example.py ✓ 1% ▏
pytest_sugar_example/test_fixture.py ✓ 2% ▎
pytest_sugar_example/test_parametrize.py ✓✓ 4% ▍
pytest_sugar_example/test_repeat_example.py ✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓ 23% ██▍
✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓ 42% ████▎
✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓ 62% ██████▎
✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓ 81% ████████▏
✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓100% ██████████
---------------------------------------------------- benchmark: 1 tests ---------------------------------------------------
Name (time in ns) Min Max Mean StdDev Median IQR Outliers OPS (Mops/s) Rounds Iterations
---------------------------------------------------------------------------------------------------------------------------
test_concat 302.8003 3,012.5000 328.2844 97.9087 321.5999 8.2495 866;2220 3.0461 90868 20
---------------------------------------------------------------------------------------------------------------------------
Legend:
Outliers: 1 Standard Deviation from Mean; 1.5 IQR (InterQuartile Range) from 1st Quartile and 3rd Quartile.
OPS: Operations Per Second, computed as 1 / Mean
Results (2.63s):
104 passed
6.13.9. pytest-steps: Share Data Between Tests¶
Have you ever wanted to use the result of one test for another test? That is when pytest_steps comes in handy.
In the code below, I use the result of sum_test as the input of average_2_nums. The argument steps_data allows me to share the data between 2 tests.
from pytest_steps import test_steps
def sum(n1, n2):
return n1 + n2
def average_2_nums(sum):
return sum / 2
def sum_test(steps_data):
res = sum(1, 3)
assert res == 4
steps_data.res = res
def perc_difference_test(steps_data):
avg = average_2_nums(steps_data.res)
assert avg == 2
@test_steps(sum_test, perc_difference_test)
def test_calc_suite(test_step, steps_data):
if test_step == 'sum_test':
sum_test(steps_data)
elif test_step == 'perc_difference_test':
perc_difference_test(steps_data)
$ pytest test_steps.py
============================= test session starts ==============================
platform darwin -- Python 3.8.10, pytest-7.1.2, pluggy-0.13.1
rootdir: /Users/khuyen/book/Efficient_Python_tricks_and_tools_for_data_scientists/Chapter5
plugins: anyio-3.5.0, steps-1.8.0, typeguard-2.12.1
collecting ...
collected 2 items
test_steps.py .. [100%]
============================== 2 passed in 0.02s ===============================
6.13.10. Pandera: a Python Library to Validate Your Pandas DataFrame¶
!pip install pandera
The outputs of your pandas DataFrame might not be like what you expected either due to the error in your code or the change in the data format. Using data that is different from what you expected can cause errors or lead to decrease performance.
Thus, it is important to validate your data before using it. A good tool to validate pandas DataFrame is pandera. Pandera is easy to read and use.
import pandera as pa
from pandera import check_input
import pandas as pd
df = pd.DataFrame({"col1": [5.0, 8.0, 10.0], "col2": ["text_1", "text_2", "text_3"]})
schema = pa.DataFrameSchema(
{
"col1": pa.Column(float, pa.Check(lambda minute: 5 <= minute)),
"col2": pa.Column(str, pa.Check.str_startswith("text_")),
}
)
validated_df = schema(df)
validated_df
| col1 | col2 | |
|---|---|---|
| 0 | 5.0 | text_1 |
| 1 | 8.0 | text_2 |
| 2 | 10.0 | text_3 |
You can also use the pandera’s decorator check_input to validates input pandas DataFrame before entering the function.
@check_input(schema)
def plus_three(df):
df["col1_plus_3"] = df["col1"] + 3
return df
plus_three(df)
| col1 | col2 | col1_plus_3 | |
|---|---|---|---|
| 0 | 5.0 | text_1 | 8.0 |
| 1 | 8.0 | text_2 | 11.0 |
| 2 | 10.0 | text_3 | 13.0 |
6.13.11. DeepDiff Find Deep Differences of Python Objects¶
!pip install deepdiff
When testing the outputs of your functions, it can be frustrated to see your tests fail because of something you don’t care too much about such as:
order of items in a list
different ways to specify the same thing such as abbreviation
exact value up to the last decimal point, etc
Is there a way that you can exclude certain parts of the object from the comparison? That is when DeepDiff comes in handy.
from deepdiff import DeepDiff
DeepDiff can output a meaningful comparison like below:
price1 = {'apple': 2, 'orange': 3, 'banana': [3, 2]}
price2 = {'apple': 2, 'orange': 3, 'banana': [2, 3]}
DeepDiff(price1, price2)
{'values_changed': {"root['banana'][0]": {'new_value': 2, 'old_value': 3},
"root['banana'][1]": {'new_value': 3, 'old_value': 2}}}
With DeepDiff, you also have full control of which characteristics of the Python object DeepDiff should ignore. In the example below, since the order is ignored [3, 2] is equivalent to [2, 3].
# Ignore orders
DeepDiff(price1, price2, ignore_order=True)
{}
We can also exclude certain part of our object from the comparison. In the code below, we ignore ml and machine learning since ml is a abbreviation of machine learning.
experience1 = {"machine learning": 2, "python": 3}
experience2 = {"ml": 2, "python": 3}
DeepDiff(
experience1,
experience2,
exclude_paths={"root['ml']", "root['machine learning']"},
)
{}
Cmpare 2 numbers up to a specific decimal point:
num1 = 0.258
num2 = 0.259
DeepDiff(num1, num2, significant_digits=2)
{}
6.13.12. hypothesis: Property-based Testing in Python¶
!pip install hypothesis
If you want to test some properties or assumptions, it can be cumbersome to write a wide range of scenarios. To automatically run your tests against a wide range of scenarios and find edge cases in your code that you would otherwise have missed, use hypothesis.
In the code below, I test if the addition of two floats is commutative. The test fails when either x or y is NaN.
# test_hypothesis.py
from hypothesis import given
from hypothesis.strategies import floats
@given(floats(), floats())
def test_floats_are_commutative(x, y):
assert x + y == y + x
$ pytest test_hypothesis.py
Test session starts (platform: linux, Python 3.8.10, pytest 6.2.5, pytest-sugar 0.9.4)
benchmark: 3.4.1 (defaults: timer=time.perf_counter disable_gc=False min_rounds=5 min_time=0.000005 max_time=1.0 calibration_precision=10 warmup=False warmup_iterations=100000)
rootdir: /home/khuyen/book/book/Chapter5
plugins: hydra-core-1.1.1, Faker-8.12.1, benchmark-3.4.1, repeat-0.9.1, anyio-3.3.0, hypothesis-6.31.6, sugar-0.9.4
collecting ...
――――――――――――――――――――――――― test_floats_are_commutative ――――――――――――――――――――――――――
@given(floats(), floats())
> def test_floats_are_commutative(x, y):
test_hypothesis.py:7:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
x = 0.0, y = nan
@given(floats(), floats())
def test_floats_are_commutative(x, y):
> assert x + y == y + x
E assert (0.0 + nan) == (nan + 0.0)
test_hypothesis.py:8: AssertionError
---------------------------------- Hypothesis ----------------------------------
Falsifying example: test_floats_are_commutative(
x=0.0, y=nan, # Saw 1 signaling NaN
)
test_hypothesis.py ⨯ 100% ██████████
=========================== short test summary info ============================
FAILED test_hypothesis.py::test_floats_are_commutative - assert (0.0 + nan) =...
Results (0.38s):
1 failed
- test_hypothesis.py:6 test_floats_are_commutative
Now I can rewrite my code to make it more robust against these edge cases.
6.13.13. Deepchecks: Check Category Mismatch Between Train and Test Set¶
!pip install deepchecks
Sometimes, it is important to know if your test set contains the same categories in the train set. If you want to check the category mismatch between the train and test set, use Deepchecks’s CategoryMismatchTrainTest.
In the example below, the result shows that there are 2 new categories in the test set. They are ‘d’ and ‘e’.
from deepchecks.checks.integrity.new_category import CategoryMismatchTrainTest
from deepchecks.base import Dataset
import pandas as pd
train = pd.DataFrame({"col1": ["a", "b", "c"]})
test = pd.DataFrame({"col1": ["c", "d", "e"]})
train_ds = Dataset(train, cat_features=["col1"])
test_ds = Dataset(test, cat_features=["col1"])
CategoryMismatchTrainTest().run(train_ds, test_ds)
Category Mismatch Train Test
Find new categories in the test set.
Additional Outputs
| Number of new categories | Percent of new categories in sample | New categories examples | |
|---|---|---|---|
| Column | |||
| col1 | 2 | 66.67% | ['d', 'e'] |