Create model with R that have a validation

voxsim · April 2012

Is there any way to create a model with extension of R?

I have to do a validation with Split-Validation, i read the code of some existing R classifier, but when I try to validate them as shown in the video tutorials, they don't return a valid object Model.

voxsim · April 2012

I merged decisiontree and execute scripts


package com.rapidminer.operator.r;

import java.net.URL;
import java.util.List;

import org.rosuda.REngine.REXP;
import org.rosuda.REngine.REXPMismatchException;
import org.rosuda.REngine.RList;

import com.rapidminer.example.Attribute;
import com.rapidminer.example.ExampleSet;
import com.rapidminer.example.Statistics;
import com.rapidminer.example.set.SplittedExampleSet;
import com.rapidminer.operator.IOObject;
import com.rapidminer.operator.Model;
import com.rapidminer.operator.OperatorCapability;
import com.rapidminer.operator.OperatorDescription;
import com.rapidminer.operator.OperatorException;
import com.rapidminer.operator.UserError;
import com.rapidminer.parameter.ParameterType;
import com.rapidminer.parameter.ParameterTypeString;
import com.rapidminer.parameter.ParameterTypeText;
import com.rapidminer.parameter.TextType;
import com.rapidminer.tools.r.RPlotPainter;
import com.rapidminer.tools.r.RSession;
import com.rapidminer.tools.r.RSessionListener;
import com.rapidminer.tools.r.RSessionManager;
import com.rapidminer.tools.r.translation.RTranslations;
import com.rapidminer.tools.r.translation.RTranslator;
import com.rapidminer.operator.learner.AbstractLearner;
import com.rapidminer.operator.learner.PredictionModel;
import com.rapidminer.operator.learner.tree.GreaterSplitCondition;
import com.rapidminer.operator.learner.tree.LessEqualsSplitCondition;
import com.rapidminer.operator.learner.tree.NominalSplitCondition;
import com.rapidminer.operator.learner.tree.SplitCondition;
import com.rapidminer.operator.learner.tree.Tree;
import com.rapidminer.operator.learner.tree.TreeModel;

public class RDecisionTreeLearner extends AbstractLearner implements RSessionListener {
	public static final String PARAMETER_R_SCRIPT = "script";
	public static final String PARAMETER_INPUT = "input";
	public static final String PARAMETER_VARIABLE_NAME = "name_of_variable";

	private transient boolean isErrorOccurred = false;
	private transient String errorOccured;

	public RDecisionTreeLearner(OperatorDescription description) {
        super(description);
    }

@Override
	public Class<? extends PredictionModel> getModelClass() {
		return TreeModel.class;
	}

	public List<ParameterType> getParameterTypes() {
		List<ParameterType> types = super.getParameterTypes();

		ParameterType type = new ParameterTypeText(PARAMETER_R_SCRIPT, "This script will be executed on one of the available R servers.", TextType.PLAIN, true);
		type.setExpert(false);
		types.add(type);
		type = new ParameterTypeString(PARAMETER_INPUT, "This assigns each input port a variable name. If the type of input object is supported by the R translation, it will be accessible under this variable name.");
		type.setExpert(false);
		types.add(type);

		return types;
	}

	public void changeTreeToLeaf(Tree node, ExampleSet exampleSet) {
        Attribute label = exampleSet.getAttributes().getLabel();
        exampleSet.recalculateAttributeStatistics(label);
        int labelValue = (int)exampleSet.getStatistics(label, Statistics.MODE);
        
        if(labelValue != -1) {
        	String labelName = label.getMapping().mapIndex(labelValue);
        	node.setLeaf(labelName);
        	for (String value : label.getMapping().getValues()) {
        		int count = (int)exampleSet.getStatistics(label, Statistics.COUNT, value);
        		node.addCount(value, count);
        	}
        }
    }

	public Tree getChilds(Tree current, Attribute bestAttribute, ExampleSet exampleSet, REXP payload, int depth) throws REXPMismatchException {
		int i = 0;
		int j = -1;
		SplittedExampleSet splitted = null;

		if(payload.isString()) {
			String[] names = payload._attr().asList().at(0).asStrings();
			String[] values = payload.asStrings();
			while (i < names.length) {
				String name = names;
				String value = values;

				if (name!=null && value!=null) {
					SplitCondition condition = null;

					if (bestAttribute.isNominal()) {
						condition = new NominalSplitCondition(bestAttribute, value);
						splitted = SplittedExampleSet.splitByAttribute(exampleSet, bestAttribute);
						j += 1;
					} else {
						double bestSplitValue = Double.valueOf(name.substring(2)).doubleValue();
						if (i == 0) {
							condition = new LessEqualsSplitCondition(bestAttribute, bestSplitValue);
							splitted = SplittedExampleSet.splitByAttribute(exampleSet, bestAttribute, bestSplitValue);
							j = 0;
						} else {
							condition = new GreaterSplitCondition(bestAttribute, bestSplitValue);
							j = 1;
						}
					}

					splitted.selectSingleSubset(j);
					ExampleSet eSet = (ExampleSet)splitted.clone();
					Tree child = new Tree(eSet);
					child.setLeaf(value);
					changeTreeToLeaf(child, eSet);
					current.addChild(child, condition);
				}
				i++;
			}
			return current;
		}

		if(payload.isList()) {
			RList rList = payload.asList();
			while (i < payload.length()) {
				String name = rList.keyAt(i);
				if (name!=null) {
					SplitCondition condition = null;

					if (bestAttribute.isNominal()) {
						condition = new NominalSplitCondition(bestAttribute, name);
						splitted = SplittedExampleSet.splitByAttribute(exampleSet, bestAttribute);
						j += 1;
					} else {
						double bestSplitValue = Double.valueOf(name.substring(2)).doubleValue();
						if (i == 0) {
							condition = new LessEqualsSplitCondition(bestAttribute, bestSplitValue);
							splitted = SplittedExampleSet.splitByAttribute(exampleSet, bestAttribute, bestSplitValue);
							j = 0;
						} else {
							condition = new GreaterSplitCondition(bestAttribute, bestSplitValue);
							j = 1;
						}
					}

					splitted.selectSingleSubset(j);
					Tree child = buildTreeFromR((ExampleSet)splitted.clone(), rList.at(i), depth);
					current.addChild(child, condition);
				}
				i++;
			}
			return current;
		}

		return current;
	}

	// Pari un nodo, dispari un arco
	public Tree buildTreeFromR(ExampleSet exampleSet, REXP payload, int depth) throws REXPMismatchException {
		Tree current = null;

		if(payload.isString() && payload.length() == 1) {
			String name = payload.asString();
			current = new Tree((ExampleSet) exampleSet.clone());
			current.setLeaf(name);
			changeTreeToLeaf(current, exampleSet);
			return current;
		}

		int i = 0;
		if(payload.isList()) {
			RList rList = payload.asList();
			while (i < payload.length()) {
				String name = rList.keyAt(i);
				if (name!=null) {
					current = new Tree((ExampleSet) exampleSet.clone());
					current.setLeaf(name);

					Attribute bestAttribute = exampleSet.getAttributes().get(name);
					current = getChilds(current, bestAttribute, exampleSet, rList.at(i), depth+1);
				}
				i++;
			}
			return current;
		}

		return current;
	}

	public Tree translateRTreeToTree(ExampleSet exampleSet) throws OperatorException {
		isErrorOccurred = false;
		errorOccured = null;

		Tree root = new Tree((ExampleSet)exampleSet.clone());
		root.setLeaf("Errore");
		/*Attribute label = exampleSet.getAttributes().get("TargetRelativo");
		exampleSet.getAttributes().setLabel(label);*/

		// try retrieving the connection
		RSession rSession = null;
		rSession = RSessionManager.acquireSession();
		rSession.registerSessionListener(this);

		try {
			// making RapidMiner Input available in R
			String inputVariableName = getParameterAsString(PARAMETER_INPUT);
			RTranslator<? extends IOObject> translator = RTranslations.getTranslators(ExampleSet.class);
			if (translator != null) {
				translator.exportObject(rSession, inputVariableName, (IOObject) exampleSet);
			} else {
				throw new UserError(this, "r.no_translator_available", ExampleSet.class.getSimpleName());
			}

			if (isParameterSet(PARAMETER_R_SCRIPT))
				rSession.execute(getParameterAsString(PARAMETER_R_SCRIPT));
			// checking for errors during execution occurred on log
			if (isErrorOccurred)
				throw new UserError(this, new Throwable(errorOccured), "r.r_error");	

			REXP payload = rSession.eval("result");
			root = buildTreeFromR(exampleSet, payload, 0);
		} catch (REXPMismatchException e) {
			e.printStackTrace();
		} finally {
			RSessionManager.releaseSession(rSession);
		}

		return root;
	}

@Override
	public Model learn(ExampleSet exampleSet) throws OperatorException {
		// learn tree
		Tree root = translateRTreeToTree(exampleSet);
		// create and return model
		return new TreeModel(exampleSet, root);
	}

@Override
	public boolean supportsCapability(OperatorCapability capability) {
		switch (capability) {
        case BINOMINAL_ATTRIBUTES:
        case POLYNOMINAL_ATTRIBUTES:
        case NUMERICAL_ATTRIBUTES:
        case POLYNOMINAL_LABEL:
        case BINOMINAL_LABEL:
        case WEIGHTED_EXAMPLES:
        case MISSING_VALUES:
            return true;
        default:
            return false;
        }
	}

@Override
	public void informErrors(String[] errors) {
		StringBuilder builder = new StringBuilder();
		for (String error : errors) {
			isErrorOccurred = true;
			builder.append(error);
			logError(error);
		}
		errorOccured = builder.toString();
	}

@Override
	public void informWarnings(String[] warnings) {
		for (String warning : warnings)
			logWarning(warning);
	}

@Override
	public void informOutput(String[] text) {
		for (String warning : text)
			logNote(warning);
	}
	/* Ignore following events */
@Override
	public void informAssignment(RSession session) {
	}

@Override
	public void informEvaluation(RSession session) {
	}

@Override
	public void informExecution(RSession session) {
	}

@Override
	public void informInterpretation(RSession session) {
	}

@Override
	public void notifyPlotListener(RPlotPainter plotPainter) {
	}

@Override
	public void informHelpChange(URL helpPage) {
	}
}

This run well .. i put the result in a variable called result. The script in R is an implementation of C4.5 and the result is an associative array.

Simon

Create model with R that have a validation

Answers

Categories