Using ML Models in a Java Client

Introduction

This guide demonstrates how to use a downloaded ML model generated using WSO2 ML in plain Java code. This guide uses the <ML_HOME>/samples/model-usage/ sample Java client implementation.

Prerequisites

Before you run this sample, do the following.

1. Download WSO2 ML, and start the server. For instructions, see Getting Started.
2. Generate a model using WSO2 ML. This model will be used to make the predictions. For instructions on generating a model, see Generating Models. To build a model, see ML UI Workflow.
3. Download the generated model.

Building the sample

To build this sample, a Java project should be created as follows.

1. Create a Maven project.

2. Include the following dependencies in the pom.xml file of the project.

   <dependency>
   	<groupId>org.wso2.carbon.ml</groupId>
   	<artifactId>org.wso2.carbon.ml.core</artifactId>
   	<version>1.0.1-SNAPSHOT</version>
   </dependency>
   <dependency>
   	<groupId>org.wso2.carbon.ml</groupId>
   	<artifactId>org.wso2.carbon.ml.commons</artifactId>
   	<version>1.0.1-SNAPSHOT</version>
   </dependency>

   Runtime dependencies 

   <!--Dependencies required for MLModel deserialization-->
   <dependency>
   	<groupId>org.wso2.orbit.org.apache.spark</groupId>
   	<artifactId>spark-core_2.10</artifactId>
   	<version>1.4.1.wso2v1</version>
   </dependency>
   <dependency>
   	<groupId>org.wso2.orbit.org.apache.spark</groupId>
   	<artifactId>spark-mllib_2.10</artifactId>
   	<version>1.4.1.wso2v1</version>
   </dependency>
   <dependency>
   	<groupId>org.scala-lang</groupId>
   	<artifactId>scala-library</artifactId>
   	<version>2.10.4</version>
   </dependency>
   <dependency>
   	<groupId>org.wso2.orbit.org.scalanlp</groupId>
   	<artifactId>breeze_2.10</artifactId>
   	<version>0.11.1.wso2v1</version>
   </dependency>
   <dependency>
   	<groupId>org.wso2.orbit.github.fommil.netlib</groupId>
   	<artifactId>core</artifactId>
   	<version>1.1.2.wso2v1</version>
   </dependency>
   <!--Dependencies required for predictions-->
   <dependency>
   	<groupId>org.wso2.orbit.sourceforge.f2j</groupId>
   	<artifactId>arpack_combined</artifactId>
   	<version>0.1.wso2v1</version>
   </dependency>

3. The downloaded model contains a serialized MLModel object. Deserialize the MLModel as follows.

   private MLModel deserializeMLModel() throws IOException, ClassNotFoundException, URISyntaxException {
     // Path to downloaded-ml-model
     String pathToDownloadedModel = “/tmp/downloaded-ml-model”;
     FileInputStream fileInputStream = new FileInputStream(pathToDownloadedModel);
     ObjectInputStream in = new ObjectInputStream(fileInputStream);
     MLModel mlModel = (MLModel) in.readObject();
   
     logger.info("Algorithm Type : " + mlModel.getAlgorithmClass());
     logger.info("Algorithm Name : " + mlModel.getAlgorithmName());
     logger.info("Response Variable : " + mlModel.getResponseVariable());
     logger.info("Features : " + mlModel.getFeatures());
     return mlModel;
   }

4. Use org.wso2.carbon.ml.core.impl.Predictor to make predictions using the downloaded ML Model. Predictor accepts a list of string arrays as feature values. The feature values should be in the order of the feature index inside the array as shown below.

   MLModelUsageSample modelUsageSample = new MLModelUsageSample();
   // Deserialize
   MLModel mlModel = modelUsageSample.deserializeMLModel();
   // Predict
   String[] featureValueArray1 = new String[] { "6", "148", "72", "35", "0", "33.6", "0.627", "50" };
   String[] featureValueArray2 = new String[] { "0", "101", "80", "40", "0", "26", "0.5", "33" };
   ArrayList<String[]> list = new ArrayList<String[]>();
   list.add(featureValueArray1);
   list.add(featureValueArray2);
   modelUsageSample.predict(list, mlModel);
   public void predict(List<String[]> featureValueLists, MLModel mlModel) throws MLModelHandlerException {
   	Predictor predictor = new Predictor(0, mlModel, featureValueLists);
   	List<?> predictions = predictor.predict();
   	logger.info("Predictions : " + predictions);
   }

The complete project created above can be found in the <ML_HOME>/samples/model-usage directory.  To build this, execute the following command from this directory.

mvn clean install

Executing the sample

To execute this sample, execute the following command from the <ML_HOME>/samples/model-usage directory.

mvn exec:java

Analysing the output

The following output is displayed in the ML console once you execute the sample.

INFO  [MLModelUsageSample] - Algorithm Type : Classification
INFO  [MLModelUsageSample] - Algorithm Name : LOGISTIC_REGRESSION
INFO  [MLModelUsageSample] - Response Variable : Class
INFO  [MLModelUsageSample] - Features : [Feature [name=Age, index=7, type=NUMERICAL, imputeOption=DISCARD, include=true], Feature [name=BMI, index=5, type=NUMERICAL, imputeOption=DISCARD, include=true], Feature [name=DBP, index=2, type=NUMERICAL, imputeOption=DISCARD, include=true], Feature [name=DPF, index=6, type=NUMERICAL, imputeOption=DISCARD, include=true], Feature [name=NumPregnancies, index=0, type=NUMERICAL, imputeOption=DISCARD, include=true], Feature [name=PG2, index=1, type=NUMERICAL, imputeOption=DISCARD, include=true], Feature [name=SI2, index=4, type=NUMERICAL, imputeOption=DISCARD, include=true], Feature [name=TSFT, index=3, type=NUMERICAL, imputeOption=DISCARD, include=true]]
Sep 03, 2015 9:35:20 AM com.github.fommil.netlib.BLAS <clinit>
WARNING: Failed to load implementation from: com.github.fommil.netlib.NativeSystemBLAS
Sep 03, 2015 9:35:20 AM com.github.fommil.netlib.BLAS <clinit>
WARNING: Failed to load implementation from: com.github.fommil.netlib.NativeRefBLAS
INFO  [MLModelUsageSample] - Predictions : [0, 0]