Home

Awesome

Deequ - Unit Tests for Data

GitHub license GitHub issues Build Status Maven Central

Deequ is a library built on top of Apache Spark for defining "unit tests for data", which measure data quality in large datasets. We are happy to receive feedback and contributions.

Python users may also be interested in PyDeequ, a Python interface for Deequ. You can find PyDeequ on GitHub, readthedocs, and PyPI.

Requirements and Installation

Deequ depends on Java 8. Deequ version 2.x only runs with Spark 3.1, and vice versa. If you rely on a previous Spark version, please use a Deequ 1.x version (legacy version is maintained in legacy-spark-3.0 branch). We provide legacy releases compatible with Apache Spark versions 2.2.x to 3.0.x. The Spark 2.2.x and 2.3.x releases depend on Scala 2.11 and the Spark 2.4.x, 3.0.x, and 3.1.x releases depend on Scala 2.12.

Available via maven central.

Choose the latest release that matches your Spark version from the available versions. Add the release as a dependency to your project. For example, for Spark 3.1.x:

Maven

<dependency>
  <groupId>com.amazon.deequ</groupId>
  <artifactId>deequ</artifactId>
  <version>2.0.0-spark-3.1</version>
</dependency>

sbt

libraryDependencies += "com.amazon.deequ" % "deequ" % "2.0.0-spark-3.1"

Example

Deequ's purpose is to "unit-test" data to find errors early, before the data gets fed to consuming systems or machine learning algorithms. In the following, we will walk you through a toy example to showcase the most basic usage of our library. An executable version of the example is available here.

Deequ works on tabular data, e.g., CSV files, database tables, logs, flattened json files, basically anything that you can fit into a Spark dataframe. For this example, we assume that we work on some kind of Item data, where every item has an id, a productName, a description, a priority and a count of how often it has been viewed.

case class Item(
  id: Long,
  productName: String,
  description: String,
  priority: String,
  numViews: Long
)

Our library is built on Apache Spark and is designed to work with very large datasets (think billions of rows) that typically live in a distributed filesystem or a data warehouse. For the sake of simplicity in this example, we just generate a few toy records though.

val rdd = spark.sparkContext.parallelize(Seq(
  Item(1, "Thingy A", "awesome thing.", "high", 0),
  Item(2, "Thingy B", "available at http://thingb.com", null, 0),
  Item(3, null, null, "low", 5),
  Item(4, "Thingy D", "checkout https://thingd.ca", "low", 10),
  Item(5, "Thingy E", null, "high", 12)))

val data = spark.createDataFrame(rdd)

Most applications that work with data have implicit assumptions about that data, e.g., that attributes have certain types, do not contain NULL values, and so on. If these assumptions are violated, your application might crash or produce wrong outputs. The idea behind deequ is to explicitly state these assumptions in the form of a "unit-test" for data, which can be verified on a piece of data at hand. If the data has errors, we can "quarantine" and fix it, before we feed it to an application.

The main entry point for defining how you expect your data to look is the VerificationSuite from which you can add Checks that define constraints on attributes of the data. In this example, we test for the following properties of our data:

In code this looks as follows:

import com.amazon.deequ.VerificationSuite
import com.amazon.deequ.checks.{Check, CheckLevel, CheckStatus}


val verificationResult = VerificationSuite()
  .onData(data)
  .addCheck(
    Check(CheckLevel.Error, "unit testing my data")
      .hasSize(_ == 5) // we expect 5 rows
      .isComplete("id") // should never be NULL
      .isUnique("id") // should not contain duplicates
      .isComplete("productName") // should never be NULL
      // should only contain the values "high" and "low"
      .isContainedIn("priority", Array("high", "low"))
      .isNonNegative("numViews") // should not contain negative values
      // at least half of the descriptions should contain a url
      .containsURL("description", _ >= 0.5)
      // half of the items should have less than 10 views
      .hasApproxQuantile("numViews", 0.5, _ <= 10))
    .run()

After calling run, deequ translates your test to a series of Spark jobs, which it executes to compute metrics on the data. Afterwards it invokes your assertion functions (e.g., _ == 5 for the size check) on these metrics to see if the constraints hold on the data. We can inspect the VerificationResult to see if the test found errors:

import com.amazon.deequ.constraints.ConstraintStatus


if (verificationResult.status == CheckStatus.Success) {
  println("The data passed the test, everything is fine!")
} else {
  println("We found errors in the data:\n")

  val resultsForAllConstraints = verificationResult.checkResults
    .flatMap { case (_, checkResult) => checkResult.constraintResults }

  resultsForAllConstraints
    .filter { _.status != ConstraintStatus.Success }
    .foreach { result => println(s"${result.constraint}: ${result.message.get}") }
}

If we run the example, we get the following output:

We found errors in the data:

CompletenessConstraint(Completeness(productName)): Value: 0.8 does not meet the requirement!
PatternConstraint(containsURL(description)): Value: 0.4 does not meet the requirement!

The test found that our assumptions are violated! Only 4 out of 5 (80%) of the values of the productName attribute are non-null and only 2 out of 5 (40%) values of the description attribute did contain a url. Fortunately, we ran a test and found the errors, somebody should immediately fix the data :)

More examples

Our library contains much more functionality than what we showed in the basic example. We are in the process of adding more examples for its advanced features. So far, we showcase the following functionality:

Citation

If you would like to reference this package in a research paper, please cite:

Sebastian Schelter, Dustin Lange, Philipp Schmidt, Meltem Celikel, Felix Biessmann, and Andreas Grafberger. 2018. Automating large-scale data quality verification. Proc. VLDB Endow. 11, 12 (August 2018), 1781-1794.

License

This library is licensed under the Apache 2.0 License.