Great Assertions

This library is inspired by the Great Expectations library. The library has made the various expectations found in Great Expectations available when using the inbuilt python unittest assertions.

Install

pip install great-assertions

Code example Pandas

from great_assertions import GreatAssertions
import pandas as pd

class GreatAssertionTests(GreatAssertions):
    def test_expect_table_row_count_to_equal(self):
        df = pd.DataFrame({"col_1": [100, 200, 300], "col_2": [10, 20, 30]})
        self.expect_table_row_count_to_equal(df, 3)

Code example PySpark

from great_assertions import GreatAssertions
from pyspark.sql import SparkSession

class GreatAssertionTests(GreatAssertions):

    def setUp(self):
        self.spark = SparkSession.builder.getOrCreate()

    def test_expect_table_row_count_to_equal(self):
        df = self.spark.createDataFrame(
            [
                {"col_1": 100, "col_2": 10},
                {"col_1": 200, "col_2": 20},
                {"col_1": 300, "col_2": 30},
            ]
        )
        self.expect_table_row_count_to_equal(df, 3)

List of available assertions

	Pandas	PySpark
expect_table_row_count_to_equal	✅	✅
expect_table_row_count_to_be_greater_than	✅	✅
expect_table_row_count_to_be_less_than	✅	✅
expect_table_has_no_duplicate_rows	✅	✅
expect_column_value_to_equal	✅	✅
expect_column_values_to_be_between	✅	✅
expect_column_values_to_match_regex	✅	✅
expect_column_values_to_be_in_set	✅	✅
expect_column_values_to_be_of_type	✅	✅
expect_table_columns_to_match_ordered_list	✅	✅
expect_table_columns_to_match_set	✅	✅
expect_date_range_to_be_more_than	✅	✅
expect_date_range_to_be_less_than	✅	✅
expect_date_range_to_be_between	✅	✅
expect_column_mean_to_be_between	✅	✅
expect_column_value_counts_percent_to_be_between	✅	✅
expect_frame_equal	✅	✅
expect_column_has_no_duplicate_rows	✅	✅
expect_column_value_to_equal_if	✅	✅
expect_column_value_to_be_greater_if	✅	✅

Assertion Descriptions

For a description of the assertions see Assertion Definitions

Running the tests

Executing the tests still require unittest, the following options have been tested with the examples provided.

Option 1

import unittest
suite = unittest.TestLoader().loadTestsFromTestCase(GreatAssertionTests)
runner = unittest.TextTestRunner(verbosity=2)
runner.run(suite) 

Options 2

if __name__ == '__main__':
    unittest.main()   

Pie Charts and Tables

For a more visual representation of the results, when using in Databricks or Jupyter Notebooks. The results can be outputted as tables or pie-chart.

import unittest
from great_assertions import GreatAssertionResult, GreatAssertions

class DisplayTest(GreatAssertions):
    def test_pass1(self):
        assert True is True

    def test_fail(self):
        assert "Hello" == "World"    

suite = unittest.TestLoader().loadTestsFromTestCase(DisplayTest)
test_runner = unittest.runner.TextTestRunner(resultclass = GreatAssertionResult)
result = test_runner.run(suite)

result.to_barh() #Also available: result.to_pie()

result.to_results_table()

result.to_full_results_table()

Runnng with XML-Runner

To run with xml-runner, there is no difference to how it's currently used. However you will not be able to get method like to_results_table as these use a different resultclass

import xmlrunner
suite = unittest.TestLoader().loadTestsFromTestCase(DisplayTest)
test_runner = xmlrunner.XMLRunner(output="test-results")
test_runner.run(suite)

Production Monitoring

The assertions provided by GA will also allow the validation of the any environment including Production. Currently GA only supports saving the results to Spark, for example databricks.

Once the run has completed there is a save method, as seen below.

import xmlrunner
suite = unittest.TestLoader().loadTestsFromTestCase(DisplayTest)
test_runner = xmlrunner.XMLRunner(output="test-results")
result = test_runner.run(suite)
result.save(format="databricks")

The image below shows a simple graph of the accumulation of tests over test run. However much more complex analysis can be performed with the extended data being generated by GA.

The extended table of results contains the following:

run_id	timestamp	method	information	test_id	status	extended
20211222093029	2021-12-22 09:30:29	test_fail8	Traceback (most recent call last...	13	Fail	{"id": 13, "name": "expect_date_range_to_be_less_than", "values": {"expected_max_date": "2019-05-13", "actual_max_date": "2019-05-13"}}
20211222093029	2021-12-22 09:30:29	test_fail9	Traceback (most recent call last...	14	Fail	{"id": 14, "name": "expect_date_range_to_be_more_than", "values": {"expected_min_date": "2015-10-01", "actual_min_date": "2015-10-01"}}

From the extended column you can get further details about the type test, which was executed and the results. For example if we look at the test expect_table_row_count_to_be_less_than we should assert that the max row should not be breached.

In the code below, the expected was 100 and the actual was 205, which caused the test to fail. Therefore Analysts can query the extended data to get a picture of the size of the breach.

extended = {
    "id": 2,
    "name": expect_table_row_count_to_be_less_than,
    "values": {
        "exp_max_count": 100,
        "act_count": 205,
    },
}

In production monitoring these types of results can allow the prevention of skewed results. For example, if you had a result, where the expected values were withing a range of 0-100 and you got an exceptionally large value.

The large value could cause business functionality to be skewed such that a defect could causes damage or loss of income or incorrect reporting to a downstream system.

Therefore, GA will allow you to provide benchmarks to the production validation and an experienced analyst can create reports on top of the data.

An example of the extended dataset:

Notes

If you get an arrows function warning when running in Databricks, this will happen because a toPandas() method is being used for many of the assertions. The plan is to remove Pandas conversion for pure PySpark code. If this is an issue, please raise an issue so this method can be prioritised. For now, it’s advisable to make sure the datasets are not too big, which cause the driver to crash.

Development

To create a development environment, create a virtualenv and make a development installation

virtualenv ve
source ve/bin/activate

To run tests, just use pytest

(ve) pytest