This project is for making a simple example of running an ML model in KSQL. Run mvn clean package to
build.
The Jupyter notebook in the notebooks/
directory generates the model file iris-pipeline.pkl.z. The jar file files/jars/jpmml-sklearn-executable-1.5-SNAPSHOT.jar
is used to convert this model file into a PMML file models/iris-pipeline.pmml.
The Docker Compose file is used to start the Confluent Platform components required to run the KSQL server.
Before running docker-compose up -d to start it, move the project jar target/ksql-ml-pmml-example-0.0.1-jar-with-dependencies.jar
and the model PMML file models/iris-pipeline.pmml into the mounted volume docker/volume.
There is a branch loadModelOnce which addresses the inefficiency of loading the model on each invocation of the UDF "predict"
method.
This code was developed for the Medium article ML in KSQL.