Cascading Pattern

How to quickly migrate Predictive Models (PMML) from SAS, R, Micro Strategies etc., onto Hadoop and deploy them at scale

Introduction

Cascading Pattern is a machine learning project within the Cascading development framework used to build enterprise data workflows. The Cascading framework provides an abstraction layer on top of Hadoop and other computing topologies. It allows enterprises to leverage existing skills and resources to build data processing applications on Apache Hadoop, without specialized Hadoop skills. Pattern, in particular, leverages an industry standard called Predictive Model Markup Language (PMML), which allows data scientists to leverage their favorite statistical & analytics tools such as R, Oracle, etc., to export predictive models and quickly run them on data sets stored in Hadoop. Pattern’s benefits include reduced development costs, time savings and reduced licensing issues at scale – all while leveraging Hadoop clusters, core competencies of analytics staff, and existing intellectual property in the predictive models.

Enterprise Use Cases

In the context of an Enterprise app, this would tend to fit in the diagram shown below. Analytics work in the back office typically is where predictive modeling gets performed — using SAS, R, SPSS, Microstrategy, Teradata, etc.

enterprise

By using Cascading Pattern, predictive modeling can now be exported as PMML from a variety of analytics frameworks, then run on Apache Hadoop at scale. This saves licensing costs while allowing for applications to scale-out and allows for predictive modeling to be integrated directly within other business logic expressed as Cascading apps.

Step 1 - Environment Setup

In this section, we will go through the steps needed to setup your environment.

  • To set up Java for your environment, download Java and follow the installation instructions.

    • Version 1.6.x was used to create the examples used here

    • Get the JDK, not the JRE

    • Install according to vendor instructions

    • Be sure to set the JAVA_HOME environment variable correctly

  • To set up Gradle for your environment, download Gradle and follow the installation instructions.

    • Version 1.4 and later is required for some examples in this tutorial

    • Install according to vendor instructions

    • Be sure to set the GRADLE_HOME environment variable correctly

  • Install a version of Apache Hadoop in local mode or if you want to have a full cluster to experiment with, use the Cascading Vagrant Hadoop Cluster.

    • Please make sure, you install a stable version of hadoop from the 1.x series and not hadoop 2.

  • Set up R and RStudio for your environment by visiting:

Step 2 - Source Code

Navigate to the Pattern Github project, to download the source code which we will be using.

In the bottom right corner of the screen, click ``Download ZIP'' to download a ZIP compressed archive of the source code for the Cascading Pattern project.

github

Once this has completed downloading, unzip and move the directory ``pattern'' to a location on your file system where you have space available to work.

Step 3 - Model Creation

Navigate to the pattern directory, and then into its pattern-examples subdirectory. There is an example R script in examples/r/rf_pmml.R that creates a Random Forest model. This is representative of a predictive model for an anti-fraud classifier used in e-commerce apps.

## train a RandomForest model
f <- as.formula("as.factor(label) ~ .")
fit <- randomForest(f, data_train, ntree=50)

## test the model on the holdout test set
print(fit$importance)
print(fit)

predicted <- predict(fit, data)
data$predicted <- predicted
confuse <- table(pred = predicted, true = data[,1])
print(confuse)

## export predicted labels to TSV
write.table(data, file=paste(dat_folder, "sample.tsv", sep="/"),
  quote=FALSE, sep="\t", row.names=FALSE)

## export RF model to PMML
saveXML(pmml(fit), file=paste(dat_folder, "sample.rf.xml", sep="/"))

Load the rf_pmml.R'' script into RStudio using the File'' menu and ``Open File..'' option.

Click the ``Source'' button in the upper middle section of the screen.

r-studio

This will execute the R script and create the predictive model. The last line saves the predictive model into a file called sample.rf.xml as PMML. If you take a look at that PMML, it’s XML and not optimal for humans to read, but efficient for machines to parse:

<?xml version="1.0"?>
<PMML version="4.0" xmlns="http://www.dmg.org/PMML-4_0"
 xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
 xsi:schemaLocation="http://www.dmg.org/PMML-4_0
 http://www.dmg.org/v4-0/pmml-4-0.xsd">
 <Header copyright="Copyright (c)2012 Concurrent, Inc."
  description="Random Forest Tree Model">
  <Extension name="user" value="ceteri" extender="Rattle/PMML"/>
  <Application name="Rattle/PMML" version="1.2.30"/>
  <Timestamp>2012-10-22 19:39:28</Timestamp>
 </Header>
 <DataDictionary numberOfFields="4">
  <DataField name="label" optype="categorical" dataType="string">
   <Value value="0"/>
   <Value value="1"/>
  </DataField>
  <DataField name="var0" optype="continuous" dataType="double"/>
  <DataField name="var1" optype="continuous" dataType="double"/>
  <DataField name="var2" optype="continuous" dataType="double"/>
 </DataDictionary>
 <MiningModel modelName="randomForest_Model" functionName="classification">
  <MiningSchema>
   <MiningField name="label" usageType="predicted"/>
   <MiningField name="var0" usageType="active"/>
   <MiningField name="var1" usageType="active"/>
   <MiningField name="var2" usageType="active"/>
  </MiningSchema>
  <Segmentation multipleModelMethod="majorityVote">
   <Segment id="1">
    <True/>
    <TreeModel modelName="randomForest_Model" functionName="classification"
     algorithmName="randomForest" splitCharacteristic="binarySplit">
     <MiningSchema>
      <MiningField name="label" usageType="predicted"/>
      <MiningField name="var0" usageType="active"/>
      <MiningField name="var1" usageType="active"/>
      <MiningField name="var2" usageType="active"/>
     </MiningSchema>
...

Cascading Pattern supports additional models, as well as ensembles of the following models.

  • General Regression

  • Regression

  • Clustering

  • Tree

  • Mining

Step 4 - Cascading Build

Now that we have a model created and exported as PMML, let’s work on running it at scale atop Apache Hadoop.

In the pattern-examples directory, execute the following Bash shell commands:

> gradle clean jar

That line invokes Gradle to run the build script build.gradle, and compile the Cascading Pattern example app.

After that compiles look for the built app as a JAR file in the build/libs subdirectory:

> ls -lts build/libs/pattern-examples-*.jar

Now we’re ready to run this Cascading Pattern example app on Apache Hadoop. First, we make sure to delete the output results (required by Hadoop). Then we run Hadoop: we specify the JAR file for the app, the PMML file using a --pmml command line option, along with sample input data data/sample.tsv and the location of the output results:

> rm -rf out
> hadoop jar build/libs/pattern-examples-*.jar \
    data/sample.tsv out/classify --pmml data/sample.rf.xml

After that runs, check the out/classify subdirectory. Look at the results of running the PMML model, which will be in the part-* partition files:

> less out/classify/part-*

Let’s take a look at what we just built and ran. The source code for this example is located in the src/main/java/cascading/pattern/Main.java file:

public class Main
  {
  /** @param args  */
  public static void main( String[] args ) throws RuntimeException
    {
    String inputPath = args[ 0 ];
    String classifyPath = args[ 1 ];

    // set up the config properties
    Properties properties = new Properties();
    AppProps.setApplicationJarClass( properties, Main.class );

    HadoopFlowConnector flowConnector = new HadoopFlowConnector( properties );

    // create source and sink taps
    Tap inputTap = new Hfs( new TextDelimited( true, "\t" ), inputPath );
    Tap classifyTap = new Hfs( new TextDelimited( true, "\t" ), classifyPath );

    // handle command line options
    OptionParser optParser = new OptionParser();
    optParser.accepts( "pmml" ).withRequiredArg();

    OptionSet options = optParser.parse( args );

    // connect the taps, pipes, etc., into a flow
    FlowDef flowDef = FlowDef.flowDef()
      .setName( "classify" )
      .addSource( "input", inputTap )
      .addSink( "classify", classifyTap );

    // build a Cascading assembly from the PMML description
    if( options.hasArgument( "pmml" ) )
      {
      String pmmlPath = (String) options.valuesOf( "pmml" ).get( 0 );

      PMMLPlanner pmmlPlanner = new PMMLPlanner()
        .setPMMLInput( new File( pmmlPath ) )
        .retainOnlyActiveIncomingFields()
        .setDefaultPredictedField( new Fields( "predict", Double.class ) );
      // default value if missing from the model

      flowDef.addAssemblyPlanner( pmmlPlanner );
      }

    // write a DOT file and run the flow
    Flow classifyFlow = flowConnector.connect( flowDef );
    classifyFlow.writeDOT( "dot/classify.dot" );
    classifyFlow.complete();
    }
  }

Most of the code is the basic plumbing used for Cascading apps. The portions which are specific to Cascading Pattern and PMML are the few lines involving the pmmlPlanner object.

Additional Resources

Back to Top - Last updated 2016-08-30 10:16:36 UTC

© 2007-2015 Concurrent, Inc. All rights reserved.