Back to Guides

Using Kogito to add prediction capabilities to an application

This guide demonstrates how your Quarkus application can use Kogito to add business automation to power it up with predictions.

Kogito is a next generation business automation toolkit that originates from the well known Open Source project Drools (for predictions). Kogito aims at providing another approach to business automation where the main message is to expose your business knowledge (processes, rules, decisions, predictions) in a domain specific way.

Prerequisites

To complete this guide, you need:

  • Roughly 15 minutes

  • An IDE

  • JDK 11+ installed with JAVA_HOME configured appropriately

  • Apache Maven 3.8.6

  • A working container runtime (Docker or Podman)

  • Optionally the Quarkus CLI if you want to use it

  • Optionally Mandrel or GraalVM installed and configured appropriately if you want to build a native executable (or Docker if you use a native container build)

Architecture

In this example, we build a very simple microservice which offers one REST endpoint:

  • /LogisticRegressionIrisData

This endpoint will be automatically generated based on given PMML file, that in turn will make use of generated code to make certain predictions based on the data being processed.

PMML file

The PMML file describes the prediction logic of our microservice. It should provide the actual model (Regression, Tree, Scorecard, Clustering, etc.) needed to make the prediction.

Prediction endpoints

Those are the entry points to the service that can be consumed by clients.

Solution

We recommend that you follow the instructions in the next sections and create the application step by step. However, you can go right to the complete example.

Clone the Git repository: git clone https://github.com/quarkusio/quarkus-quickstarts.git, or download an archive.

The solution is located in the kogito-pmml-quickstart directory.

Creating the Maven Project

First, we need a new project. Create a new project with the following command:

CLI
quarkus create app org.acme:kogito-pmml-quickstart \
    --extension='kogito,resteasy-reactive-jackson' \
    --no-code
cd kogito-pmml-quickstart

To create a Gradle project, add the --gradle or --gradle-kotlin-dsl option.

For more information about how to install the Quarkus CLI and use it, please refer to the Quarkus CLI guide.

Maven
mvn io.quarkus.platform:quarkus-maven-plugin:2.16.0.Final:create \
    -DprojectGroupId=org.acme \
    -DprojectArtifactId=kogito-pmml-quickstart \
    -Dextensions='kogito,resteasy-reactive-jackson' \
    -DnoCode
cd kogito-pmml-quickstart

To create a Gradle project, add the -DbuildTool=gradle or -DbuildTool=gradle-kotlin-dsl option.

This command generates a Maven project, importing the kogito extension that comes with all needed dependencies and configuration to equip your application with business automation. It also imports the resteasy-reactive-jackson extension that is needed for Kogito to expose REST services.

If you already have your Quarkus project configured, you can add the kogito extension to your project by running the following command in your project base directory:

CLI
quarkus extension add 'kogito'
Maven
./mvnw quarkus:add-extension -Dextensions='kogito'
Gradle
./gradlew addExtension --extensions='kogito'

This will add the following to your build file:

pom.xml
<dependency>
    <groupId>org.kie.kogito</groupId>
    <artifactId>kogito-quarkus</artifactId>
</dependency>
build.gradle
implementation("org.kie.kogito:kogito-quarkus")

Writing the application

Predictions are evaluated based on a PMML model, whose standard and specifications may be read here. Let’s start by adding a simple PMML file: LogisticRegressionIrisData.pmml. It contains a Regression model named LogisticRegressionIrisData, and it uses a regression function to predict plant species from sepal and petal dimensions:

<PMML xmlns="http://www.dmg.org/PMML-4_2" version="4.2">
  <Header/>
  <DataDictionary numberOfFields="5">
    <DataField name="Sepal.Length" optype="continuous" dataType="double"/>
    <DataField name="Sepal.Width" optype="continuous" dataType="double"/>
    <DataField name="Petal.Length" optype="continuous" dataType="double"/>
    <DataField name="Petal.Width" optype="continuous" dataType="double"/>
    <DataField name="Species" optype="categorical" dataType="string">
      <Value value="setosa"/>
      <Value value="virginica"/>
      <Value value="versicolor"/>
    </DataField>
  </DataDictionary>
  <RegressionModel functionName="classification" modelName="LogisticRegressionIrisData" targetFieldName="Species">
    <MiningSchema>
      <MiningField name="Sepal.Length"/>
      <MiningField name="Sepal.Width"/>
      <MiningField name="Petal.Length"/>
      <MiningField name="Petal.Width"/>
      <MiningField name="Species" usageType="target"/>
    </MiningSchema>
    <Output>
      <OutputField name="Probability_setosa" optype="continuous" dataType="double" feature="probability" value="setosa"/>
      <OutputField name="Probability_versicolor" optype="continuous" dataType="double" feature="probability" value="versicolor"/>
      <OutputField name="Probability_virginica" optype="continuous" dataType="double" feature="probability" value="virginica"/>
    </Output>
    <RegressionTable targetCategory="setosa" intercept="0.11822288946815">
      <NumericPredictor name="Sepal.Length" exponent="1" coefficient="0.0660297693761902"/>
      <NumericPredictor name="Sepal.Width" exponent="1" coefficient="0.242847872054487"/>
      <NumericPredictor name="Petal.Length" exponent="1" coefficient="-0.224657116235727"/>
      <NumericPredictor name="Petal.Width" exponent="1" coefficient="-0.0574727291860025"/>
    </RegressionTable>
    <RegressionTable targetCategory="versicolor" intercept="1.57705897385745">
      <NumericPredictor name="Sepal.Length" exponent="1" coefficient="-0.0201536848255179"/>
      <NumericPredictor name="Sepal.Width" exponent="1" coefficient="-0.44561625761404"/>
      <NumericPredictor name="Petal.Length" exponent="1" coefficient="0.22066920522933"/>
      <NumericPredictor name="Petal.Width" exponent="1" coefficient="-0.494306595747785"/>
    </RegressionTable>
    <RegressionTable targetCategory="virginica" intercept="-0.695281863325603">
      <NumericPredictor name="Sepal.Length" exponent="1" coefficient="-0.0458760845506725"/>
      <NumericPredictor name="Sepal.Width" exponent="1" coefficient="0.202768385559553"/>
      <NumericPredictor name="Petal.Length" exponent="1" coefficient="0.00398791100639665"/>
      <NumericPredictor name="Petal.Width" exponent="1" coefficient="0.551779324933787"/>
    </RegressionTable>
  </RegressionModel>
</PMML>

During project compilation, Kogito will read the file and generate the classes needed for the evaluation, together with a couple of REST endpoints.

To get started quickly copy the PMML file from the quickstart.

Running and Using the Application

Running in Dev Mode

To run the microservice in dev mode, use:

CLI
quarkus dev
Maven
./mvnw quarkus:dev
Gradle
./gradlew --console=plain quarkusDev

Running in JVM Mode

When you’re done playing with dev mode you can run it as a standard Java application.

First compile it:

CLI
quarkus build
Maven
./mvnw install
Gradle
./gradlew build

Then run it:

java -jar target/quarkus-app/quarkus-run.jar

Running in Native Mode

This same demo can be compiled into native code: no modifications required.

This implies that you no longer need to install a JVM on your production environment, as the runtime technology is included in the produced binary, and optimized to run with minimal resource overhead.

Compilation will take a bit longer, so this step is disabled by default; let’s build a native executable with the following command:

CLI
quarkus build --native
Maven
./mvnw install -Dnative
Gradle
./gradlew build -Dquarkus.package.type=native

After getting a cup of coffee, you’ll be able to run this binary directly:

./target/kogito-pmml-quickstart-1.0.0-SNAPSHOT-runner

Testing the Application

To test your application, just send a request to the service with giving the person as JSON payload.

curl -X POST http://localhost:8080/LogisticRegressionIrisData \
    -H 'content-type: application/json' \
    -H 'accept: application/json' \
    -d '{ "Sepal.Length": 6.9, "Sepal.Width": 3.1, "Petal.Length": 5.1, "Petal.Width": 2.3 }'

In the response, you should see the prediction, that should be virginica:

{
  "Species": "virginica"
}

You can also invoke the descriptive endpoint, that will provide also the OutputField evaluated:

curl -X POST http://localhost:8080/LogisticRegressionIrisData/descriptive \
    -H 'content-type: application/json' \
    -H 'accept: application/json' \
    -d '{ "Sepal.Length": 6.9, "Sepal.Width": 3.1, "Petal.Length": 5.1, "Petal.Width": 2.3 }'
{
    "correlationId": null,
    "segmentationId": null,
    "segmentId": null,
    "segmentIndex": 0,
    "resultCode": "OK",
    "resultObjectName": "Species",
    "resultVariables": {
        "Probability_setosa": 0.04871813160275851,
        "Probability_versicolor": 0.04509592640753013,
        "Probability_virginica": 0.9061859419897114,
        "Species": "virginica"
    }
}