The Intelligent Decisions journey with DMN and PMML

This repository contains an end to end demonstration that starts with some data, trains a machine learning model and deploys it with Kogito.

Requirements

The following tools must be installed on the system:

• Docker >= 20.10.3
• Docker-compose >= 1.25.2
• Java >= 11
• Maven >= 3.6.3
• R >= 3.6.3

The following R packages must be installed: pmml, randomForest.

Step 1 - Generation of the training set

Generate the training set using the script generate_dataset.R

Rscript generate_dataset.R

A file dataset.csv will be created under the current directory.

Step 2 - train the random forest model

Train and export the PMML model using the script train.R

Rscript train.R

A file risk_rf.pmml will be created under the current directory.

Step 3 - create a DMN model that adds some logic to the machine learning model

Start the business central container with

docker run -it -p8080:8080 --rm jboss/business-central-workbench-showcase:7.51.0.Final

Follow the next instructions:

• Open http://localhost:8080/business-central/kie-wb.jsp with your browser.
• Login with admin:admin.
• Add a new namespace mySpace and a new project myMortgage.
• Click import asset and select the pmml model you created in the previous step. When it is opened, replace the pmml version from 4.4 to 4.2. Save and download the model.
• Add another asset with the file myMortgage.dmn and open it. Click on the Included Models section and select the risk_rf pmml model, then open the Risk Score Model bkm function and set the risk_rf model and inputs. Save and download the model

Step 4 - Create the kogito application

Create a new quarkus kogito application with

mvn io.quarkus:quarkus-maven-plugin:2.0.3.Final:create \
    -DprojectGroupId=org.acme \
    -DprojectArtifactId=kogito-quickstart \
    -Dextensions="kogito" \
    -DnoExamples
cd kogito-quickstart

and add the following dependencies in the pom.xml file

    <!-- PMML -->
    <dependency>
      <groupId>org.kie.kogito</groupId>
      <artifactId>kogito-addons-quarkus-tracing-decision</artifactId>
    </dependency>
    <dependency>
      <groupId>org.kie.kogito</groupId>
      <artifactId>kogito-pmml</artifactId>
    </dependency>
    <dependency>
      <groupId>org.kie</groupId>
      <artifactId>kie-dmn-jpmml</artifactId>
      <version>7.55.0.Final</version>
    </dependency>
    <dependency>
      <groupId>org.jpmml</groupId>
      <artifactId>pmml-evaluator</artifactId>
      <version>1.5.15</version>
    </dependency>
    <dependency>
      <groupId>org.jpmml</groupId>
      <artifactId>pmml-evaluator-extension</artifactId>
      <version>1.5.15</version>
    </dependency>
    <dependency>
      <groupId>io.quarkus</groupId>
      <artifactId>quarkus-smallrye-openapi</artifactId>
    </dependency>
    <dependency>
      <groupId>org.kie.kogito</groupId>
      <artifactId>kogito-addons-quarkus-monitoring-prometheus</artifactId>
    </dependency>

Add the following property in the application.properties file so to get the nice swagger-ui

quarkus.swagger-ui.always-include=true

and copy the myMortgage.dmn and risk_rf.pmml files under the resources folder of the project.

Package the application with

mvn clean package -DskipTests

and build the docker image with

docker build -f src/main/docker/Dockerfile.jvm -t quay.io/jrota/pmml-kogito:1.0 .

Step 5 - Deploy the kogito application with the trustyAI infra

Clone the kogito-examples repository and checkout the release branch 1.9.0.Final

git clone https://github.com/kiegroup/kogito-examples.git
git checkout 1.9.0.Final

open the file kogito-examples/trusty-demonstration/docker-compose/docker-compose.yml and replace the kogito-app image with quay.io/jrota/pmml-kogito:1.0.

Stop the business central container so to free the 8080 port. Start docker-compose with

docker-compose -f kogito-examples/trusty-demonstration/docker-compose/docker-compose.yml up

Copy the generated grafana dashboards to the docker-compose/grafana folder

cp target/classes/META-INF/resources/monitoring/dashboards/* kogito-examples/trusty-demonstration/docker-compose/grafana/provisioning/dashboards/

Step 6 - execute some requests and check the trusty console

Open the kogito application swagger-ui at localhost:8080/q/swagger-ui and send a POST request to the endpoint myMortgage with the following payload

{
  "Finantial Situation": {
    "MonthlySalary": 2000,
    "TotalAsset": 10000
  },
  "Mortgage Request": {
    "TotalRequired": 100000,
    "NumberInstallments": 100
  },
  "Applicant": {
    "First Name": "Jacopo",
    "Last Name": "Rota",
    "Age": 29,
    "Email": "[email protected]"
  }
}

Check the trusty audit console at localhost:1338 and play with it!