cuallee

Meaning good in Aztec (Nahuatl), pronounced: QUAL-E

This library provides an intuitive API to describe data quality checks initially just for PySpark dataframes v3.3.0. And extended to pandas, snowpark, duckdb, daft and more. It is a replacement written in pure python of the pydeequ framework.

I gave up in deequ as after extensive use, the API is not user-friendly, the Python Callback servers produce additional costs in our compute clusters, and the lack of support to the newest version of PySpark.

As result cuallee was born

This implementation goes in hand with the latest API from PySpark and uses the Observation API to collect metrics at the lower cost of computation. When benchmarking against pydeequ, cuallee uses circa <3k java classes underneath and remarkably less memory.

Support

cuallee is the data quality framework truly dataframe agnostic.

Provider	API	Versions
	snowpark	1.11.1, 1.4.0
	pyspark & spark-connect	3.5.x, 3.4.0, 3.3.x, 3.2.x
	bigquery	3.4.1
	pandas	2.0.2, 1.5.x, 1.4.x
	duckdb	1.0.0, 0.10.2,0.9.2,0.8.0
	polars	1.0.0, 0.19.6
	daft	0.2.24, 0.2.19

_{Logos are trademarks of their own brands.}

Install

pip install cuallee

Checks

The most common checks for data integrity validations are completeness and uniqueness an example of this dimensions shown below:

from cuallee import Check, CheckLevel # WARN:0, ERR: 1

# Nulls on column Id
check = Check(CheckLevel.WARNING, "Completeness")
(
    check
    .is_complete("id")
    .is_unique("id")
    .validate(df)
).show() # Returns a pyspark.sql.DataFrame

Dates

Perhaps one of the most useful features of cuallee is its extensive number of checks for Date and Timestamp values. Including, validation of ranges, set operations like inclusion, or even a verification that confirms continuity on dates using the is_daily check function.

# Unique values on id
check = Check(CheckLevel.WARNING, "CheckIsBetweenDates")
df = spark.sql(
    """
    SELECT
        explode(
            sequence(
                to_date('2022-01-01'),
                to_date('2022-01-10'),
                interval 1 day)) as date
    """)
assert (
    check.is_between("date", ("2022-01-01", "2022-01-10"))
    .validate(df)
    .first()
    .status == "PASS"
)

Membership

Other common test is the validation of list of values as part of the multiple integrity checks required for better quality data.

df = spark.createDataFrame([[1, 10], [2, 15], [3, 17]], ["ID", "value"])
check = Check(CheckLevel.WARNING, "is_contained_in_number_test")
check.is_contained_in("value", (10, 15, 20, 25)).validate(df)

Regular Expressions

When it comes to the flexibility of matching, regular expressions are always to the rescue. cuallee makes use of the regular expressions to validate that fields of type String conform to specific patterns.

df = spark.createDataFrame([[1, "is_blue"], [2, "has_hat"], [3, "is_smart"]], ["ID", "desc"])
check = Check(CheckLevel.WARNING, "has_pattern_test")
check.has_pattern("desc", r"^is.*t$") # only match is_smart 33% of rows.
check.validate(df).first().status == "FAIL"

Anomalies

Statistical tests are a great aid for verifying anomalies on data. Here an example that shows that will PASS only when 40% of data is inside the interquartile range

df = spark.range(10)
check = Check(CheckLevel.WARNING, "IQR_Test")
check.is_inside_interquartile_range("id", pct=0.4)
check.validate(df).first().status == "PASS"

+---+-------------------+-----+-------+------+-----------------------------+-----+----+----------+---------+--------------+------+
|id |timestamp          |check|level  |column|rule                         |value|rows|violations|pass_rate|pass_threshold|status|
+---+-------------------+-----+-------+------+-----------------------------+-----+----+----------+---------+--------------+------+
|1  |2022-10-19 00:09:39|IQR  |WARNING|id    |is_inside_interquartile_range|10000|10  |4         |0.6      |0.4           |PASS  |
+---+-------------------+-----+-------+------+-----------------------------+-----+----+----------+---------+--------------+------+

Workflows (Process Mining)

Besides the common citizen-like checks, cuallee offers out-of-the-box real-life checks. For example, suppose that you are working SalesForce or SAP environment. Very likely your business processes will be driven by a lifecycle:

• Order-To-Cash
• Request-To-Pay
• Inventory-Logistics-Delivery
• Others. In this scenario, cuallee offers the ability that the sequence of events registered over time, are according to a sequence of events, like the example below:

import pyspark.sql.functions as F
from cuallee import Check, CheckLevel

data = pd.DataFrame({
   "name":["herminio", "herminio", "virginie", "virginie"],
   "event":["new","active", "new", "active"],
   "date": ["2022-01-01", "2022-01-02", "2022-01-03", "2022-02-04"]}
   )
df = spark.createDataFrame(data).withColumn("date", F.to_date("date"))

# Cuallee Process Mining
# Testing that all edges on workflows
check = Check(CheckLevel.WARNING, "WorkflowViolations")

# Validate that 50% of data goes from new => active
check.has_workflow("name", "event", "date", [("new", "active")], pct=0.5)
check.validate(df).show(truncate=False)

+---+-------------------+------------------+-------+-------------------------+------------+--------------------+----+----------+---------+--------------+------+
|id |timestamp          |check             |level  |column                   |rule        |value               |rows|violations|pass_rate|pass_threshold|status|
+---+-------------------+------------------+-------+-------------------------+------------+--------------------+----+----------+---------+--------------+------+
|1  |2022-11-07 23:08:50|WorkflowViolations|WARNING|('name', 'event', 'date')|has_workflow|(('new', 'active'),)|4   |2.0       |0.5      |0.5           |PASS  |
+---+-------------------+------------------+-------+-------------------------+------------+--------------------+----+----------+---------+--------------+------+

Assertions

[2024-09-28] ✨ New feature! Return a simple true|false as a unified result for your check

import pandas as pd
from cuallee import Check
df = pd.DataFrame({"X":[1,2,3]})
# .ok(dataframe) method of a check will call validate and then verify that all rules are PASS
assert Check().is_complete("X").ok(df)

Controls

Simplify the entire validation of a dataframe in a particular dimension.

import pandas as pd
from cuallee import Control
df = pd.DataFrame({"X":[1,2,3], "Y": [10,20,30]})
# Checks all columns in dataframe for using is_complete check
Control.completeness(df)

cuallee VS pydeequ

In the test folder there are docker containers with the requirements to match the tests. Also a perftest.py available at the root folder for interests.

# 1000 rules / # of seconds

cuallee: ▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 162.00
pydeequ: ▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 322.00

Catalogue

Check	Description	DataType
is_complete	Zero nulls	agnostic
is_unique	Zero duplicates	agnostic
is_primary_key	Zero duplicates	agnostic
are_complete	Zero nulls on group of columns	agnostic
are_unique	Composite primary key check	agnostic
is_composite_key	Zero duplicates on multiple columns	agnostic
is_greater_than	col > x	numeric
is_positive	col > 0	numeric
is_negative	col < 0	numeric
is_greater_or_equal_than	col >= x	numeric
is_less_than	col < x	numeric
is_less_or_equal_than	col <= x	numeric
is_equal_than	col == x	numeric
is_contained_in	col in [a, b, c, ...]	agnostic
is_in	Alias of is_contained_in	agnostic
not_contained_in	col not in [a, b, c, ...]	agnostic
not_in	Alias of not_contained_in	agnostic
is_between	a <= col <= b	numeric, date
has_pattern	Matching a pattern defined as a regex	string
is_legit	String not null & not empty ^\S$	string
has_min	min(col) == x	numeric
has_max	max(col) == x	numeric
has_std	σ(col) == x	numeric
has_mean	μ(col) == x	numeric
has_sum	Σ(col) == x	numeric
has_percentile	%(col) == x	numeric
has_cardinality	count(distinct(col)) == x	agnostic
has_infogain	count(distinct(col)) > 1	agnostic
has_max_by	A utilitary predicate for max(col_a) == x for max(col_b)	agnostic
has_min_by	A utilitary predicate for min(col_a) == x for min(col_b)	agnostic
has_correlation	Finds correlation between 0..1 on corr(col_a, col_b)	numeric
has_entropy	Calculates the entropy of a column entropy(col) == x for classification problems	numeric
is_inside_interquartile_range	Verifies column values reside inside limits of interquartile range Q1 <= col <= Q3 used on anomalies.	numeric
is_in_millions	col >= 1e6	numeric
is_in_billions	col >= 1e9	numeric
is_t_minus_1	For date fields confirms 1 day ago t-1	date
is_t_minus_2	For date fields confirms 2 days ago t-2	date
is_t_minus_3	For date fields confirms 3 days ago t-3	date
is_t_minus_n	For date fields confirms n days ago t-n	date
is_today	For date fields confirms day is current date t-0	date
is_yesterday	For date fields confirms 1 day ago t-1	date
is_on_weekday	For date fields confirms day is between Mon-Fri	date
is_on_weekend	For date fields confirms day is between Sat-Sun	date
is_on_monday	For date fields confirms day is Mon	date
is_on_tuesday	For date fields confirms day is Tue	date
is_on_wednesday	For date fields confirms day is Wed	date
is_on_thursday	For date fields confirms day is Thu	date
is_on_friday	For date fields confirms day is Fri	date
is_on_saturday	For date fields confirms day is Sat	date
is_on_sunday	For date fields confirms day is Sun	date
is_on_schedule	For date fields confirms time windows i.e. 9:00 - 17:00	timestamp
is_daily	Can verify daily continuity on date fields by default. [2,3,4,5,6] which represents Mon-Fri in PySpark. However new schedules can be used for custom date continuity	date
has_workflow	Adjacency matrix validation on 3-column graph, based on group, event, order columns.	agnostic
is_custom	User-defined custom function applied to dataframe for row-based validation.	agnostic
satisfies	An open SQL expression builder to construct custom checks	agnostic
validate	The ultimate transformation of a check with a dataframe input for validation	agnostic

Controls pyspark

Check	Description	DataType
completeness	Zero nulls	agnostic
information	Zero nulls and cardinality > 1	agnostic
intelligence	Zero nulls, zero empty strings and cardinality > 1	agnostic
percentage_fill	% rows not empty	agnostic
percentage_empty	% rows empty	agnostic

ISO Standard

A new module has been incorporated in cuallee==0.4.0 which allows the verification of International Standard Organization columns in data frames. Simply access the check.iso interface to add the set of checks as shown below.

Check	Description	DataType
iso_4217	currency compliant ccy	string
iso_3166	country compliant country	string

df = spark.createDataFrame([[1, "USD"], [2, "MXN"], [3, "CAD"], [4, "EUR"], [5, "CHF"]], ["id", "ccy"])
check = Check(CheckLevel.WARNING, "ISO Compliant")
check.iso.iso_4217("ccy")
check.validate(df).show()
+---+-------------------+-------------+-------+------+---------------+--------------------+----+----------+---------+--------------+------+
| id|          timestamp|        check|  level|column|           rule|               value|rows|violations|pass_rate|pass_threshold|status|
+---+-------------------+-------------+-------+------+---------------+--------------------+----+----------+---------+--------------+------+
|  1|2023-05-14 18:28:02|ISO Compliant|WARNING|   ccy|is_contained_in|{'BHD', 'CRC', 'M...|   5|       0.0|      1.0|           1.0|  PASS|
+---+-------------------+-------------+-------+------+---------------+--------------------+----+----------+---------+--------------+------+

Snowflake Connection

In order to establish a connection to your SnowFlake account cuallee relies in the following environment variables to be avaialble in your environment:

• SF_ACCOUNT
• SF_USER
• SF_PASSWORD
• SF_ROLE
• SF_WAREHOUSE
• SF_DATABASE
• SF_SCHEMA

Spark Connect

Just add the environment variable SPARK_REMOTE to your remote session, then cuallee will connect using

spark_connect = SparkSession.builder.remote(os.getenv("SPARK_REMOTE")).getOrCreate()

and convert all checks to select as opposed to Observation API compute instructions.

Databricks Connection

By default cuallee will search for a SparkSession available in the globals so there is literally no need to SparkSession.builder. When working in a local environment it will automatically search for an available session, or start one.

DuckDB

For testing on duckdb simply pass your table name to your check et voilà

import duckdb
conn = duckdb.connect(":memory:")
check = Check(CheckLevel.WARNING, "DuckDB", table_name="temp/taxi/*.parquet")
check.is_complete("VendorID")
check.is_complete("tpep_pickup_datetime")
check.validate(conn)

   id            timestamp check    level                column         rule value      rows  violations  pass_rate  pass_threshold status
0   1  2022-10-31 23:15:06  test  WARNING              VendorID  is_complete   N/A  19817583         0.0        1.0             1.0   PASS
1   2  2022-10-31 23:15:06  test  WARNING  tpep_pickup_datetime  is_complete   N/A  19817583         0.0        1.0             1.0   PASS

Roadmap

100% data frame agnostic implementation of data quality checks. Define once, run everywhere

• [x] PySpark 3.5.0
• [x] PySpark 3.4.0
• [x] PySpark 3.3.0
• [x] PySpark 3.2.x
• [x] Snowpark DataFrame
• [x] Pandas DataFrame
• [x] DuckDB Tables
• [x] BigQuery Client
• [x] Polars DataFrame
• [*] Dagster Integration
• [x] Spark Connect
• [x] Daft
• [-] PDF Report
• Metadata check
• Help us in a discussion?

Whilst expanding the functionality feels a bit as an overkill because you most likely can connect spark via its drivers to whatever DBMS of your choice. In the desire to make it even more user-friendly we are aiming to make cuallee portable to all the providers above.

Authors

• canimus / Herminio Vazquez / 🇲🇽
• vestalisvirginis / Virginie Grosboillot / 🇫🇷

Contributors

Guidelines

• Contributing to cuallee

Documentation

• https://canimus.github.io/cuallee/

Paper

cuallee has been published in the Journal of Open Source Software

Vazquez et al., (2024). cuallee: A Python package for data quality checks across multiple DataFrame APIs. Journal of Open Source Software, 9(98), 6684, https://doi.org/10.21105/joss.06684

If you use cuallee please consider citing this work. Citation

License

Apache License 2.0 Free for commercial use, modification, distribution, patent use, private use. Just preserve the copyright and license.

Made with ❤️ in Utrecht 🇳🇱
Maintained over ⌛ from Ljubljana 🇸🇮
Extended 🚀 by contributions all over the 🌎