psdq 1.0.0

Ad Hoc Data Quality Tool for PySpark

PySparkDQ - Ad Hoc data quality tool for native pyspark

This library aims to be an easy to use tool to perform data quality checks at scale with PySpark. It does not require you to setup contexts nor make use of configuration files. Just instantiate the class and put some tests on your target dataframe.

Prerequisites

Before you begin, ensure you have met the following requirements:

• You have access to a Pyspark enabled environment
• Know what tests you want to run

Installation

This package is distributed on pypi. To install it just run:

pip install psdq

Quickstart

Instantiate the class with the desired Spark DataFrame and SparkSession and you're good to go.

import logging 

from pyspark.sql import SparkSession
import pyspark.sql.functions as f
from psdq import PySparkDQ

spark = SparkSession \
    .builder\
    .appName("PySparkDQ")
    .getOrCreate()

df_test = spark.createDataFrame([
                    {"name": "Alex", "birthday": '1990-03-31'},
                    {"name": "Alice", "birthday": '1990-12-09'},
                    {"name": "Bob", "birthday": '1985-06-12'},
                    {"name": "Cecilia", "birthday": '1990-12-09'},
                    {"name": "Eve", "birthday": '1964-07-22'}]) \
                    .withColumn('birthday', f.col('birthday').cast('date'))


# You can optionally add a logger to the class to check information about the execution
logging.basicConfig()
logger = logging.getLogger("PySparkDQ")
logger.setLevel(logging.INFO)

# Instantiate the test suite 
dq = PySparkDQ(spark_session=spark, df=df_test, logger=logger)

# For custom tests, create a column expression that returns bool
my_custom_test_col_expression = ((f.col('name') == 'Bob') & (f.col('birthday') == '1985-06-12'))

tests = dq\
    .values_not_null(colname="name")\
    .values_between(colname='birthday', lower_value='1990-01-01', upper_value='1993-01-01', 
                tolerance=0.5, # You can control tolerances with the optional parameters
                over_under_tolerance='over', # tolerate for Over OR Under 50% of results
                inclusive_exclusive='inclusive') # Inclusive or exclusive on exact 50% \
    .values_custom_dq(test="my custom test", # Custom test name
        partial=my_custom_test_col_expression) # Custom test col expression

Now that you have your tests, you can do one of the following:

1. Get a summary for the test run

test_results = tests.get_summary() # returns a dataframe

test_results.show()

colname	test	scope	found	total	percentage	tolerance	over_under_tolerance	inclusive_exclusive	pass
name	values_not_null	null	5	5	1	1	over	inclusive	True
birthday	values_between	1990-01-01 - 1993-01-01	3	5	0.6	0.5	over	inclusive	True
N/A	my custom test	N/A	1	5	0.2	1	over	inclusive	False

2. Thrown an exception whenever it has a failed test. in this case it will throw the following error.

tests.evaluate()

AssertionError: Detected failed tests. Count: 1, Tests: [{'colname': 'N/A', 'test': 'my custom test', 'scope': 'N/A'}]

---

3. Get a row-based evaluation of your rules. That returns the dataframe with extra folumns indicating how many tests failed and which ones failed per row

row_level_qa = tests.get_row_level_qa() # Returns a dataframe

# Returns two extra columns
# One counts failed tests per row
# Second one is a list<struct> that returns which tests have failed

# ATTENTION! Be careful when comparing this view with the summary. Tests can fail at row level and pass at summary level due to defined tolerances!
row_level_qa.show()

birthday	name	pysparkdq_fail_count	pysparkdq_failed_tests
1990-03-31	Alex	1	[Row(colname='N/A', test='my custom test', scope="Column<'((name = Bob) AND (birthday = 1985-06-12))'>")]
1990-12-09	Alice	1	[Row(colname='N/A', test='my custom test', scope="Column<'((name = Bob) AND (birthday = 1985-06-12))'>")]
1985-06-12	Bob	0	[]
1990-12-09	Cecilia	1	[Row(colname='N/A', test='my custom test', scope="Column<'((name = Bob) AND (birthday = 1985-06-12))'>")]
1964-07-22	Eve	2	[Row(colname='birthday', test='values_between', scope='1980-01-01 - 1993-01-01'), Row(colname='N/A', test='my custom test', scope="Column<'((name = Bob) AND (birthday = 1985-06-12))'>")]

Available Functions

Tests

Reference values for tests can be whatever is accepted when working with spark columns. For example, a test can compare a given column against an arbitrary value or even another column or column-expression result. Accepted values/types respect the same rules as default pyspark.

Test	Description
values_in_list	Test column for values in list, same as c.isIn(val_list)
values_not_in_list	Test column for values not in list, same as ~c.isIn(val_list)
values_null	Test column for Null values, same as c.isNull()
values_not_null	Test column for Not Null values, same as c.isNotNull()
values_equal	Test column for values equal to value, same as c == value
values_not_equal	Test column for values equal to value, same as c == value
values_between	Test column for values between boundaries, same as c.between(lower, upper)
values_greater_equal_than	Test column for values greater equal than the reference value, same as c >= value
values_lower_equal_than	Test column for values lower equal than the reference value, same as c <= value
values_greater_than	Test column for values greater than the reference value, same as c <= value
values_lower_than	Test column for values lower than the reference value, same as c <= value
values_custom_dq	Adds a custom data quality check. Needs a column expression that returns bool.

Commands

Name	Description
get_summary	Evaluates all tests and returns a SparkDataFrame with the test summaries
get_row_level_qa	Evaluates all tests row-by-row and returns the initial dataframe with extra columns for test results
evaluate	Evaluates the test summary and throws an exception if anything fails.

Contributing

If there's any issues or improvements to be requested fell free to open an issue on the board.