Home

Awesome

JPMML-Evaluator Build Status

Java Evaluator API for Predictive Model Markup Language (PMML).

Table of Contents

Features

JPMML-Evaluator is de facto the reference implementation of the PMML specification versions 3.0, 3.1, 3.2, 4.0, 4.1, 4.2, 4.3 and 4.4 for the Java/JVM platform:

For more information please see the features.md file.

JPMML-Evaluator is interoperable with most popular statistics and data mining software:

JPMML-Evaluator is fast and memory efficient. It can deliver one million scorings per second already on a desktop computer.

Prerequisites

Installation

JPMML-Evaluator library JAR files (together with accompanying Java source and Javadocs JAR files) are released via Maven Central Repository.

The current version is 1.6.6 (13 October, 2024).

The main component of JPMML-Evaluator is org.jpmml:pmml-evaluator. However, in most application scenarios, this component is not included directly, but via a data format-specific runtime component(s) org.jpmml:pmml-evaluator-${runtime} that handle the loading and storage of PMML class model objects.

The recommended data format for PMML documents is XML, and the recommended implementation is Jakarta XML Binding via the Glassfish Metro JAXB runtime:

<dependency>
	<groupId>org.jpmml</groupId>
	<artifactId>pmml-evaluator-metro</artifactId>
	<version>1.6.6</version>
</dependency>

Available components:

ComponentData format(s)
org.jpmml:pmml-evaluatorJava serialization
org.jpmml:pmml-evaluator-jacksonJSON, YAML, TOML etc. via the FasterXML Jackson suite
org.jpmml:pmml-evaluator-kryoKryo serialization
org.jpmml:pmml-evaluator-metroXML via the GlassFish Metro JAXB runtime
org.jpmml:pmml-evaluator-moxyJSON and XML via the EclipseLink MOXy JAXB runtime

API

Core types:

Core methods:

Target value types:

Target value methods:

Exception types:

Basic usage

// Building a model evaluator from a PMML file
Evaluator evaluator = new LoadingModelEvaluatorBuilder()
	.load(new File("model.pmml"))
	.build();

// Perforing the self-check
evaluator.verify();

// Printing input (x1, x2, .., xn) fields
List<InputField> inputFields = evaluator.getInputFields();
System.out.println("Input fields: " + inputFields);

// Printing primary result (y) field(s)
List<TargetField> targetFields = evaluator.getTargetFields();
System.out.println("Target field(s): " + targetFields);

// Printing secondary result (eg. probability(y), decision(y)) fields
List<OutputField> outputFields = evaluator.getOutputFields();
System.out.println("Output fields: " + outputFields);

// Iterating through columnar data (eg. a CSV file, an SQL result set)
while(true){
	// Reading a record from the data source
	Map<String, ?> arguments = readRecord();
	if(arguments == null){
		break;
	}

	// Evaluating the model
	Map<String, ?> results = evaluator.evaluate(arguments);

	// Decoupling results from the JPMML-Evaluator runtime environment
	results = EvaluatorUtil.decodeAll(results);

	// Writing a record to the data sink
	writeRecord(results);
}

// Making the model evaluator eligible for garbage collection
evaluator = null;

Advanced usage

Loading models

The PMML standard defines large number of model types. The evaluation logic for each model type is encapsulated into a corresponding ModelEvaluator subclass.

Even though ModelEvaluator subclasses can be instantiated and configured directly, the recommended approach is to follow the Builder design pattern as implemented by the ModelEvaluatorBuilder builder class.

A model evaluator builder provides configuration and loading services.

The default configuration corresponds to most common needs. It can be overriden to customize the behaviour of model evaluators for more specific needs. A model evaluator is given a copy of the configuration that was effective when the ModelEvaluatorBuilder#build() method was invoked. It is not affected by later configuration changes.

For example, creating two differently configured model evaluators from a PMML instance:

import org.jpmml.evaluator.reporting.ReportingValueFactoryFactory

PMML pmml = ...;

ModelEvaluatorBuilder modelEvaluatorBuilder = new ModelEvaluatorBuilder(pmml);

Evaluator evaluator = modelEvaluatorBuilder.build();

// Activate the generation of MathML prediction reports
modelEvaluatorBuilder.setValueFactoryFactory(ReportingValueFactoryFactory.newInstance());

Evaluator reportingEvaluator = modelEvaluatorBuilder.build();

Configurations and model evaluators are fairly lightweight, which makes them cheap to create and destroy. However, for maximum performance, it is advisable to maintain a one-to-one mapping between PMML, ModelEvaluatorBuilder and ModelEvaluator instances (ie. an application should load a PMML byte stream or file exactly once, and then maintain and reuse the resulting model evaluator as long as needed).

Some ModelEvaluator subclasses contain static caches that are lazily populated on a PMML instance basis. This may cause the first ModelEvaluator#evaluate(Map<String, ?>) method invocation to take somewhat longer to complete (relative to all the subsequent method invocations). If the model contains model verification data, then this "warm-up cost" is paid once and for all during the initial ModelEvaluator#verify() method invocation.

Thread safety

The ModelEvaluatorBuilder base class is thread safe. It is permitted to construct and configure a central ModelEvaluatorBuilder instance, and invoke its ModelEvaluatorBuilder#build() method concurrently.

Some ModelEvaluatorBuilder subclasses may extend the base class with functionality that is not thread safe. The case in point are all sorts of "loading" implementations, which modify the value of ModelEvaluatorBuilder#pmml and/or ModelEvaluatorBuilder#model fields.

The ModelEvaluator base class and all its subclasses are completely thread safe. It is permitted to share a central ModelEvaluator instance between any number of threads, and invoke its ModelEvaluator#evaluate(Map<String, ?>) method concurrently.

The JPMML-Evaluator library follow functional programming principles. In a multi-threaded environment, its data throughput capabilities should scale linearly with respect to the number of threads.

Querying the "data schema" of models

The model evaluator can be queried for the list of input (ie. independent), target (ie. primary dependent) and output (ie. secondary dependent) field definitions, which provide information about field name, data type, operational type, value domain etc.

Querying and analyzing input fields:

List<? extends InputField> inputFields = evaluator.getInputFields();
for(InputField inputField : inputFields){
	org.dmg.pmml.DataField pmmlDataField = (org.dmg.pmml.DataField)inputField.getField();
	org.dmg.pmml.MiningField pmmlMiningField = inputField.getMiningField();

	org.dmg.pmml.DataType dataType = inputField.getDataType();
	org.dmg.pmml.OpType opType = inputField.getOpType();

	switch(opType){
		case CONTINUOUS:
			com.google.common.collect.RangeSet<Double> validInputRanges = inputField.getContinuousDomain();
			break;
		case CATEGORICAL:
		case ORDINAL:
			List<?> validInputValues = inputField.getDiscreteDomain();
			break;
		default:
			break;
	}
}

Querying and analyzing target fields:

List<? extends TargetField> targetFields = evaluator.getTargetFields();
for(TargetField targetField : targetFields){
	org.dmg.pmml.DataField pmmlDataField = targetField.getField();
	org.dmg.pmml.MiningField pmmlMiningField = targetField.getMiningField(); // Could be null
	org.dmg.pmml.Target pmmlTarget = targetField.getTarget(); // Could be null

	org.dmg.pmml.DataType dataType = targetField.getDataType();
	org.dmg.pmml.OpType opType = targetField.getOpType();

	switch(opType){
		case CONTINUOUS:
			break;
		case CATEGORICAL:
		case ORDINAL:
			List<?> validTargetValues = targetField.getDiscreteDomain();

			// The list of target category values for querying HasCategoricalResults subinterfaces (HasProbability, HasConfidence etc).
			// The default element type is String.
			// If the PMML instance is pre-parsed, then the element type changes to the appropriate Java primitive type
			List<?> categories = targetField.getCategories();
			break;
		default:
			break;
	}
}

Querying and analyzing output fields:

List<? extends OutputField> outputFields = evaluator.getOutputFields();
for(OutputField outputField : outputFields){
	org.dmg.pmml.OutputField pmmlOutputField = outputField.getOutputField();

	org.dmg.pmml.DataType dataType = outputField.getDataType(); // Could be null
	org.dmg.pmml.OpType opType = outputField.getOpType(); // Could be null

	boolean finalResult = outputField.isFinalResult();
	if(!finalResult){
		continue;
	}
}

Evaluating models

A model may contain verification data, which is a small but representative set of data records (inputs plus expected outputs) for ensuring that the model evaluator is behaving correctly in this deployment configuration (JPMML-Evaluator version, Java/JVM version and vendor etc. variables). The model evaluator should be verified once, before putting it into actual use.

Performing the self-check:

evaluator.verify();

During scoring, the application code should iterate over data records (eg. rows of a table), and apply the following encode-evaluate-decode sequence of operations to each one of them.

The processing of the first data record will be significantly slower than the processing of all subsequent data records, because the model evaluator needs to lookup, validate and pre-parse model content. If the model contains verification data, then this warm-up cost is borne during the self-check.

Preparing the argument map:

Map<String, ?> inputDataRecord = ...;

Map<String, FieldValue> arguments = new LinkedHashMap<>();

List<InputField> inputFields = evaluator.getInputFields();
for(InputField inputField : inputFields){
	String inputName = inputField.getName();

	Object rawValue = inputDataRecord.get(inputName);

	// Transforming an arbitrary user-supplied value to a known-good PMML value
	// The user-supplied value is passed through: 1) outlier treatment, 2) missing value treatment, 3) invalid value treatment and 4) type conversion
	FieldValue inputValue = inputField.prepare(rawValue);

	arguments.put(inputName, inputValue);
}

Performing the evaluation:

Map<String, ?> results = evaluator.evaluate(arguments);

Extracting primary results from the result map:

List<TargetField> targetFields = evaluator.getTargetFields();
for(TargetField targetField : targetFields){
	String targetName = targetField.getName();

	Object targetValue = results.get(targetName);
}

The target value is either a Java primitive value (as a wrapper object) or a complex value as a Computable instance.

A complex target value may expose additional information about the prediction by implementing appropriate ResultFeature subinterfaces:

// Test for "entityId" result feature
if(targetValue instanceof HasEntityId){
	HasEntityId hasEntityId = (HasEntityId)targetValue;

	HasEntityRegistry<?> hasEntityRegistry = (HasEntityRegistry<?>)evaluator;
	BiMap<String, ? extends Entity> entities = hasEntityRegistry.getEntityRegistry();

	Entity winner = entities.get(hasEntityId.getEntityId());
}

// Test for "probability" result feature
if(targetValue instanceof HasProbability){
	HasProbability hasProbability = (HasProbability)targetValue;

	Set<?> categories = hasProbability.getCategories();
	for(Object category : categories){
		Double categoryProbability = hasProbability.getProbability(category);
	}
}

A complex target value may hold a reference to the model evaluator that created it. It is adisable to decode it to a Java primitive value (ie. decoupling from the JPMML-Evaluator runtime environment) as soon as all the additional information has been retrieved:

if(targetValue instanceof Computable){
	Computable computable = (Computable)targetValue;

	targetValue = computable.getResult();
}

Extracting secondary results from the result map:

List<OutputField> outputFields = evaluator.getOutputFields();
for(OutputField outputField : outputFields){
	String outputName = outputField.getName();

	Object outputValue = results.get(outputName);
}

The output value is always a Java primitive value (as a wrapper object).

Example applications

Module pmml-evaluator-example exemplifies the use of the JPMML-Evaluator library.

This module can be built using Apache Maven:

mvn clean install

The resulting uber-JAR file target/pmml-evaluator-example-executable-1.6-SNAPSHOT.jar contains the following command-line applications:

Evaluating model model.pmml with data records from input.csv. The predictions are stored to output.csv:

java -cp target/pmml-evaluator-example-executable-1.6-SNAPSHOT.jar org.jpmml.evaluator.example.EvaluationExample --model model.pmml --input input.csv --output output.csv

Evaluating model model.pmml with data records from input.csv. The predictions are verified against data records from expected-output.csv:

java -cp target/pmml-evaluator-example-executable-1.6-SNAPSHOT.jar org.jpmml.evaluator.example.TestingExample --model model.pmml --input input.csv --expected-output expected-output.csv

Enhancing model model.pmml with verification data records from input_expected-output.csv:

java -cp target/pmml-evaluator-example-executable-1.6-SNAPSHOT.jar org.jpmml.evaluator.example.EnhancementExample --model model.pmml --verification input_expected_output.csv

Getting help:

java -cp target/pmml-evaluator-example-executable-1.6-SNAPSHOT.jar <application class name> --help

Documentation

Up-to-date:

Slightly outdated:

Support

Limited public support is available via the JPMML mailing list.

License

JPMML-Evaluator is licensed under the terms and conditions of the GNU Affero General Public License, Version 3.0. For a quick summary of your rights ("Can") and obligations ("Cannot" and "Must") under AGPLv3, please refer to TLDRLegal.

If you would like to use JPMML-Evaluator in a proprietary software project, then it is possible to enter into a licensing agreement which makes JPMML-Evaluator available under the terms and conditions of the BSD 3-Clause License instead.

Additional information

JPMML-Evaluator is developed and maintained by Openscoring OÜ, Estonia.

Interested in using Java PMML API software in your company? Please contact info@openscoring.io