/* * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation; either version 2 of the License, or * (at your option) any later version. * * This program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License * along with this program; if not, write to the Free Software * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ /* * CheckAttributeSelection.java * Copyright (C) 2006 University of Waikato, Hamilton, New Zealand * */ package weka.attributeSelection; import weka.core.Attribute; import weka.core.CheckScheme; import weka.core.FastVector; import weka.core.Instances; import weka.core.MultiInstanceCapabilitiesHandler; import weka.core.Option; import weka.core.OptionHandler; import weka.core.RevisionUtils; import weka.core.SerializationHelper; import weka.core.SerializedObject; import weka.core.TestInstances; import weka.core.Utils; import weka.core.WeightedInstancesHandler; import java.util.Enumeration; import java.util.Random; import java.util.Vector; /** * Class for examining the capabilities and finding problems with * attribute selection schemes. If you implement an attribute selection using * the WEKA.libraries, you should run the checks on it to ensure robustness * and correct operation. Passing all the tests of this object does not mean * bugs in the attribute selection don't exist, but this will help find some * common ones.

* * Typical usage:

* java weka.attributeSelection.CheckAttributeSelection -W ASscheme_name * -- ASscheme_options

* * CheckAttributeSelection reports on the following: *

* Running CheckAttributeSelection with the debug option set will output the * training dataset for any failed tests.

* * The weka.attributeSelection.AbstractAttributeSelectionTest * uses this class to test all the schemes. Any changes here, have to be * checked in that abstract test class, too.

* * Valid options are:

* *

 -D
 *  Turn on debugging output.
* *
 -S
 *  Silent mode - prints nothing to stdout.
* *
 -N <num>
 *  The number of instances in the datasets (default 20).
* *
 -nominal <num>
 *  The number of nominal attributes (default 2).
* *
 -nominal-values <num>
 *  The number of values for nominal attributes (default 1).
* *
 -numeric <num>
 *  The number of numeric attributes (default 1).
* *
 -string <num>
 *  The number of string attributes (default 1).
* *
 -date <num>
 *  The number of date attributes (default 1).
* *
 -relational <num>
 *  The number of relational attributes (default 1).
* *
 -num-instances-relational <num>
 *  The number of instances in relational/bag attributes (default 10).
* *
 -words <comma-separated-list>
 *  The words to use in string attributes.
* *
 -word-separators <chars>
 *  The word separators to use in string attributes.
* *
 -eval name [options]
 *  Full name and options of the evaluator analyzed.
 *  eg: weka.attributeSelection.CfsSubsetEval
* *
 -search name [options]
 *  Full name and options of the search method analyzed.
 *  eg: weka.attributeSelection.Ranker
* *
 -test <eval|search>
 *  The scheme to test, either the evaluator or the search method.
 *  (Default: eval)
* *
 
 * Options specific to evaluator weka.attributeSelection.CfsSubsetEval:
 * 
* *
 -M
 *  Treat missing values as a seperate value.
* *
 -L
 *  Don't include locally predictive attributes.
* *
 
 * Options specific to search method weka.attributeSelection.Ranker:
 * 
* *
 -P <start set>
 *  Specify a starting set of attributes.
 *  Eg. 1,3,5-7.
 *  Any starting attributes specified are
 *  ignored during the ranking.
* *
 -T <threshold>
 *  Specify a theshold by which attributes
 *  may be discarded from the ranking.
* *
 -N <num to select>
 *  Specify number of attributes to select
* * * @author Len Trigg (trigg@cs.waikato.ac.nz) * @author FracPete (fracpete at waikato dot ac dot nz) * @version $Revision: 4783 $ * @see TestInstances */ public class CheckAttributeSelection extends CheckScheme { /* * Note about test methods: * - methods return array of booleans * - first index: success or not * - second index: acceptable or not (e.g., Exception is OK) * * FracPete (fracpete at waikato dot ac dot nz) */ /*** The evaluator to be examined */ protected ASEvaluation m_Evaluator = new CfsSubsetEval(); /*** The search method to be used */ protected ASSearch m_Search = new Ranker(); /** whether to test the evaluator (default) or the search method */ protected boolean m_TestEvaluator = true; /** * Returns an enumeration describing the available options. * * @return an enumeration of all the available options. */ public Enumeration listOptions() { Vector result = new Vector(); Enumeration en = super.listOptions(); while (en.hasMoreElements()) result.addElement(en.nextElement()); result.addElement(new Option( "\tFull name and options of the evaluator analyzed.\n" +"\teg: weka.attributeSelection.CfsSubsetEval", "eval", 1, "-eval name [options]")); result.addElement(new Option( "\tFull name and options of the search method analyzed.\n" +"\teg: weka.attributeSelection.Ranker", "search", 1, "-search name [options]")); result.addElement(new Option( "\tThe scheme to test, either the evaluator or the search method.\n" +"\t(Default: eval)", "test", 1, "-test ")); if ((m_Evaluator != null) && (m_Evaluator instanceof OptionHandler)) { result.addElement(new Option("", "", 0, "\nOptions specific to evaluator " + m_Evaluator.getClass().getName() + ":")); Enumeration enm = ((OptionHandler) m_Evaluator).listOptions(); while (enm.hasMoreElements()) result.addElement(enm.nextElement()); } if ((m_Search != null) && (m_Search instanceof OptionHandler)) { result.addElement(new Option("", "", 0, "\nOptions specific to search method " + m_Search.getClass().getName() + ":")); Enumeration enm = ((OptionHandler) m_Search).listOptions(); while (enm.hasMoreElements()) result.addElement(enm.nextElement()); } return result.elements(); } /** * Parses a given list of options.

* * Valid options are:

* *

 -D
   *  Turn on debugging output.
* *
 -S
   *  Silent mode - prints nothing to stdout.
* *
 -N <num>
   *  The number of instances in the datasets (default 20).
* *
 -nominal <num>
   *  The number of nominal attributes (default 2).
* *
 -nominal-values <num>
   *  The number of values for nominal attributes (default 1).
* *
 -numeric <num>
   *  The number of numeric attributes (default 1).
* *
 -string <num>
   *  The number of string attributes (default 1).
* *
 -date <num>
   *  The number of date attributes (default 1).
* *
 -relational <num>
   *  The number of relational attributes (default 1).
* *
 -num-instances-relational <num>
   *  The number of instances in relational/bag attributes (default 10).
* *
 -words <comma-separated-list>
   *  The words to use in string attributes.
* *
 -word-separators <chars>
   *  The word separators to use in string attributes.
* *
 -eval name [options]
   *  Full name and options of the evaluator analyzed.
   *  eg: weka.attributeSelection.CfsSubsetEval
* *
 -search name [options]
   *  Full name and options of the search method analyzed.
   *  eg: weka.attributeSelection.Ranker
* *
 -test <eval|search>
   *  The scheme to test, either the evaluator or the search method.
   *  (Default: eval)
* *
 
   * Options specific to evaluator weka.attributeSelection.CfsSubsetEval:
   * 
* *
 -M
   *  Treat missing values as a seperate value.
* *
 -L
   *  Don't include locally predictive attributes.
* *
 
   * Options specific to search method weka.attributeSelection.Ranker:
   * 
* *
 -P <start set>
   *  Specify a starting set of attributes.
   *  Eg. 1,3,5-7.
   *  Any starting attributes specified are
   *  ignored during the ranking.
* *
 -T <threshold>
   *  Specify a theshold by which attributes
   *  may be discarded from the ranking.
* *
 -N <num to select>
   *  Specify number of attributes to select
* * * @param options the list of options as an array of strings * @throws Exception if an option is not supported */ public void setOptions(String[] options) throws Exception { String tmpStr; String[] tmpOptions; super.setOptions(options); tmpStr = Utils.getOption("eval", options); tmpOptions = Utils.splitOptions(tmpStr); if (tmpOptions.length != 0) { tmpStr = tmpOptions[0]; tmpOptions[0] = ""; setEvaluator( (ASEvaluation) forName( "weka.attributeSelection", ASEvaluation.class, tmpStr, tmpOptions)); } tmpStr = Utils.getOption("search", options); tmpOptions = Utils.splitOptions(tmpStr); if (tmpOptions.length != 0) { tmpStr = tmpOptions[0]; tmpOptions[0] = ""; setSearch( (ASSearch) forName( "weka.attributeSelection", ASSearch.class, tmpStr, tmpOptions)); } tmpStr = Utils.getOption("test", options); setTestEvaluator(!tmpStr.equalsIgnoreCase("search")); } /** * Gets the current settings of the CheckAttributeSelection. * * @return an array of strings suitable for passing to setOptions */ public String[] getOptions() { Vector result; String[] options; int i; result = new Vector(); options = super.getOptions(); for (i = 0; i < options.length; i++) result.add(options[i]); result.add("-eval"); if (getEvaluator() instanceof OptionHandler) result.add( getEvaluator().getClass().getName() + " " + Utils.joinOptions(((OptionHandler) getEvaluator()).getOptions())); else result.add( getEvaluator().getClass().getName()); result.add("-search"); if (getSearch() instanceof OptionHandler) result.add( getSearch().getClass().getName() + " " + Utils.joinOptions(((OptionHandler) getSearch()).getOptions())); else result.add( getSearch().getClass().getName()); result.add("-test"); if (getTestEvaluator()) result.add("eval"); else result.add("search"); return (String[]) result.toArray(new String[result.size()]); } /** * Begin the tests, reporting results to System.out */ public void doTests() { if (getTestObject() == null) { println("\n=== No scheme set ==="); return; } println("\n=== Check on scheme: " + getTestObject().getClass().getName() + " ===\n"); // Start tests m_ClasspathProblems = false; println("--> Checking for interfaces"); canTakeOptions(); boolean weightedInstancesHandler = weightedInstancesHandler()[0]; boolean multiInstanceHandler = multiInstanceHandler()[0]; println("--> Scheme tests"); declaresSerialVersionUID(); testsPerClassType(Attribute.NOMINAL, weightedInstancesHandler, multiInstanceHandler); testsPerClassType(Attribute.NUMERIC, weightedInstancesHandler, multiInstanceHandler); testsPerClassType(Attribute.DATE, weightedInstancesHandler, multiInstanceHandler); testsPerClassType(Attribute.STRING, weightedInstancesHandler, multiInstanceHandler); testsPerClassType(Attribute.RELATIONAL, weightedInstancesHandler, multiInstanceHandler); } /** * Set the evaluator to test. * * @param value the evaluator to use. */ public void setEvaluator(ASEvaluation value) { m_Evaluator = value; } /** * Get the current evaluator * * @return the current evaluator */ public ASEvaluation getEvaluator() { return m_Evaluator; } /** * Set the search method to test. * * @param value the search method to use. */ public void setSearch(ASSearch value) { m_Search = value; } /** * Get the current search method * * @return the current search method */ public ASSearch getSearch() { return m_Search; } /** * Sets whether the evaluator or the search method is being tested. * * @param value if true then the evaluator will be tested */ public void setTestEvaluator(boolean value) { m_TestEvaluator = value; } /** * Gets whether the evaluator is being tested or the search method. * * @return true if the evaluator is being tested */ public boolean getTestEvaluator() { return m_TestEvaluator; } /** * returns either the evaluator or the search method. * * @return the object to be tested * @see #m_TestEvaluator */ protected Object getTestObject() { if (getTestEvaluator()) return getEvaluator(); else return getSearch(); } /** * returns deep copies of the given object * * @param obj the object to copy * @param num the number of copies * @return the deep copies * @throws Exception if copying fails */ protected Object[] makeCopies(Object obj, int num) throws Exception { if (obj == null) throw new Exception("No object set"); Object[] objs = new Object[num]; SerializedObject so = new SerializedObject(obj); for(int i = 0; i < objs.length; i++) { objs[i] = so.getObject(); } return objs; } /** * Performs a attribute selection with the given search and evaluation scheme * on the provided data. The generated AttributeSelection object is returned. * * @param search the search scheme to use * @param eval the evaluator to use * @param data the data to work on * @return the used attribute selection object * @throws Exception if the attribute selection fails */ protected AttributeSelection search(ASSearch search, ASEvaluation eval, Instances data) throws Exception { AttributeSelection result; result = new AttributeSelection(); result.setSeed(42); result.setSearch(search); result.setEvaluator(eval); result.SelectAttributes(data); return result; } /** * Run a battery of tests for a given class attribute type * * @param classType true if the class attribute should be numeric * @param weighted true if the scheme says it handles weights * @param multiInstance true if the scheme handles multi-instance data */ protected void testsPerClassType(int classType, boolean weighted, boolean multiInstance) { boolean PNom = canPredict(true, false, false, false, false, multiInstance, classType)[0]; boolean PNum = canPredict(false, true, false, false, false, multiInstance, classType)[0]; boolean PStr = canPredict(false, false, true, false, false, multiInstance, classType)[0]; boolean PDat = canPredict(false, false, false, true, false, multiInstance, classType)[0]; boolean PRel; if (!multiInstance) PRel = canPredict(false, false, false, false, true, multiInstance, classType)[0]; else PRel = false; if (PNom || PNum || PStr || PDat || PRel) { if (weighted) instanceWeights(PNom, PNum, PStr, PDat, PRel, multiInstance, classType); if (classType == Attribute.NOMINAL) canHandleNClasses(PNom, PNum, PStr, PDat, PRel, multiInstance, 4); if (!multiInstance) { canHandleClassAsNthAttribute(PNom, PNum, PStr, PDat, PRel, multiInstance, classType, 0); canHandleClassAsNthAttribute(PNom, PNum, PStr, PDat, PRel, multiInstance, classType, 1); } canHandleZeroTraining(PNom, PNum, PStr, PDat, PRel, multiInstance, classType); boolean handleMissingPredictors = canHandleMissing(PNom, PNum, PStr, PDat, PRel, multiInstance, classType, true, false, 20)[0]; if (handleMissingPredictors) canHandleMissing(PNom, PNum, PStr, PDat, PRel, multiInstance, classType, true, false, 100); boolean handleMissingClass = canHandleMissing(PNom, PNum, PStr, PDat, PRel, multiInstance, classType, false, true, 20)[0]; if (handleMissingClass) canHandleMissing(PNom, PNum, PStr, PDat, PRel, multiInstance, classType, false, true, 100); correctSearchInitialisation(PNom, PNum, PStr, PDat, PRel, multiInstance, classType); datasetIntegrity(PNom, PNum, PStr, PDat, PRel, multiInstance, classType, handleMissingPredictors, handleMissingClass); } } /** * Checks whether the scheme can take command line options. * * @return index 0 is true if the scheme can take options */ protected boolean[] canTakeOptions() { boolean[] result = new boolean[2]; print("options..."); if (getTestObject() instanceof OptionHandler) { println("yes"); if (m_Debug) { println("\n=== Full report ==="); Enumeration enu = ((OptionHandler) getTestObject()).listOptions(); while (enu.hasMoreElements()) { Option option = (Option) enu.nextElement(); print(option.synopsis() + "\n" + option.description() + "\n"); } println("\n"); } result[0] = true; } else { println("no"); result[0] = false; } return result; } /** * Checks whether the scheme says it can handle instance weights. * * @return true if the scheme handles instance weights */ protected boolean[] weightedInstancesHandler() { boolean[] result = new boolean[2]; print("weighted instances scheme..."); if (getTestObject() instanceof WeightedInstancesHandler) { println("yes"); result[0] = true; } else { println("no"); result[0] = false; } return result; } /** * Checks whether the scheme handles multi-instance data. * * @return true if the scheme handles multi-instance data */ protected boolean[] multiInstanceHandler() { boolean[] result = new boolean[2]; print("multi-instance scheme..."); if (getTestObject() instanceof MultiInstanceCapabilitiesHandler) { println("yes"); result[0] = true; } else { println("no"); result[0] = false; } return result; } /** * tests for a serialVersionUID. Fails in case the schemes don't declare * a UID (both must!). * * @return index 0 is true if the scheme declares a UID */ protected boolean[] declaresSerialVersionUID() { boolean[] result = new boolean[2]; boolean eval; boolean search; print("serialVersionUID..."); eval = !SerializationHelper.needsUID(m_Evaluator.getClass()); search = !SerializationHelper.needsUID(m_Search.getClass()); result[0] = eval && search; if (result[0]) println("yes"); else println("no"); return result; } /** * Checks basic prediction of the scheme, for simple non-troublesome * datasets. * * @param nominalPredictor if true use nominal predictor attributes * @param numericPredictor if true use numeric predictor attributes * @param stringPredictor if true use string predictor attributes * @param datePredictor if true use date predictor attributes * @param relationalPredictor if true use relational predictor attributes * @param multiInstance whether multi-instance is needed * @param classType the class type (NOMINAL, NUMERIC, etc.) * @return index 0 is true if the test was passed, index 1 is true if test * was acceptable */ protected boolean[] canPredict( boolean nominalPredictor, boolean numericPredictor, boolean stringPredictor, boolean datePredictor, boolean relationalPredictor, boolean multiInstance, int classType) { print("basic predict"); printAttributeSummary( nominalPredictor, numericPredictor, stringPredictor, datePredictor, relationalPredictor, multiInstance, classType); print("..."); FastVector accepts = new FastVector(); accepts.addElement("unary"); accepts.addElement("binary"); accepts.addElement("nominal"); accepts.addElement("numeric"); accepts.addElement("string"); accepts.addElement("date"); accepts.addElement("relational"); accepts.addElement("multi-instance"); accepts.addElement("not in classpath"); int numTrain = getNumInstances(), numClasses = 2, missingLevel = 0; boolean predictorMissing = false, classMissing = false; return runBasicTest(nominalPredictor, numericPredictor, stringPredictor, datePredictor, relationalPredictor, multiInstance, classType, missingLevel, predictorMissing, classMissing, numTrain, numClasses, accepts); } /** * Checks whether nominal schemes can handle more than two classes. * If a scheme is only designed for two-class problems it should * throw an appropriate exception for multi-class problems. * * @param nominalPredictor if true use nominal predictor attributes * @param numericPredictor if true use numeric predictor attributes * @param stringPredictor if true use string predictor attributes * @param datePredictor if true use date predictor attributes * @param relationalPredictor if true use relational predictor attributes * @param multiInstance whether multi-instance is needed * @param numClasses the number of classes to test * @return index 0 is true if the test was passed, index 1 is true if test * was acceptable */ protected boolean[] canHandleNClasses( boolean nominalPredictor, boolean numericPredictor, boolean stringPredictor, boolean datePredictor, boolean relationalPredictor, boolean multiInstance, int numClasses) { print("more than two class problems"); printAttributeSummary( nominalPredictor, numericPredictor, stringPredictor, datePredictor, relationalPredictor, multiInstance, Attribute.NOMINAL); print("..."); FastVector accepts = new FastVector(); accepts.addElement("number"); accepts.addElement("class"); int numTrain = getNumInstances(), missingLevel = 0; boolean predictorMissing = false, classMissing = false; return runBasicTest(nominalPredictor, numericPredictor, stringPredictor, datePredictor, relationalPredictor, multiInstance, Attribute.NOMINAL, missingLevel, predictorMissing, classMissing, numTrain, numClasses, accepts); } /** * Checks whether the scheme can handle class attributes as Nth attribute. * * @param nominalPredictor if true use nominal predictor attributes * @param numericPredictor if true use numeric predictor attributes * @param stringPredictor if true use string predictor attributes * @param datePredictor if true use date predictor attributes * @param relationalPredictor if true use relational predictor attributes * @param multiInstance whether multi-instance is needed * @param classType the class type (NUMERIC, NOMINAL, etc.) * @param classIndex the index of the class attribute (0-based, -1 means last attribute) * @return index 0 is true if the test was passed, index 1 is true if test * was acceptable * @see TestInstances#CLASS_IS_LAST */ protected boolean[] canHandleClassAsNthAttribute( boolean nominalPredictor, boolean numericPredictor, boolean stringPredictor, boolean datePredictor, boolean relationalPredictor, boolean multiInstance, int classType, int classIndex) { if (classIndex == TestInstances.CLASS_IS_LAST) print("class attribute as last attribute"); else print("class attribute as " + (classIndex + 1) + ". attribute"); printAttributeSummary( nominalPredictor, numericPredictor, stringPredictor, datePredictor, relationalPredictor, multiInstance, classType); print("..."); FastVector accepts = new FastVector(); int numTrain = getNumInstances(), numClasses = 2, missingLevel = 0; boolean predictorMissing = false, classMissing = false; return runBasicTest(nominalPredictor, numericPredictor, stringPredictor, datePredictor, relationalPredictor, multiInstance, classType, classIndex, missingLevel, predictorMissing, classMissing, numTrain, numClasses, accepts); } /** * Checks whether the scheme can handle zero training instances. * * @param nominalPredictor if true use nominal predictor attributes * @param numericPredictor if true use numeric predictor attributes * @param stringPredictor if true use string predictor attributes * @param datePredictor if true use date predictor attributes * @param relationalPredictor if true use relational predictor attributes * @param multiInstance whether multi-instance is needed * @param classType the class type (NUMERIC, NOMINAL, etc.) * @return index 0 is true if the test was passed, index 1 is true if test * was acceptable */ protected boolean[] canHandleZeroTraining( boolean nominalPredictor, boolean numericPredictor, boolean stringPredictor, boolean datePredictor, boolean relationalPredictor, boolean multiInstance, int classType) { print("handle zero training instances"); printAttributeSummary( nominalPredictor, numericPredictor, stringPredictor, datePredictor, relationalPredictor, multiInstance, classType); print("..."); FastVector accepts = new FastVector(); accepts.addElement("train"); accepts.addElement("value"); int numTrain = 0, numClasses = 2, missingLevel = 0; boolean predictorMissing = false, classMissing = false; return runBasicTest( nominalPredictor, numericPredictor, stringPredictor, datePredictor, relationalPredictor, multiInstance, classType, missingLevel, predictorMissing, classMissing, numTrain, numClasses, accepts); } /** * Checks whether the scheme correctly initialises models when * ASSearch.search is called. This test calls search with * one training dataset. ASSearch is then called on a training set with * different structure, and then again with the original training set. * If the equals method of the ASEvaluation class returns false, this is * noted as incorrect search initialisation. * * @param nominalPredictor if true use nominal predictor attributes * @param numericPredictor if true use numeric predictor attributes * @param stringPredictor if true use string predictor attributes * @param datePredictor if true use date predictor attributes * @param relationalPredictor if true use relational predictor attributes * @param multiInstance whether multi-instance is needed * @param classType the class type (NUMERIC, NOMINAL, etc.) * @return index 0 is true if the test was passed, index 1 is always false */ protected boolean[] correctSearchInitialisation( boolean nominalPredictor, boolean numericPredictor, boolean stringPredictor, boolean datePredictor, boolean relationalPredictor, boolean multiInstance, int classType) { boolean[] result = new boolean[2]; print("correct initialisation during search"); printAttributeSummary( nominalPredictor, numericPredictor, stringPredictor, datePredictor, relationalPredictor, multiInstance, classType); print("..."); int numTrain = getNumInstances(), numClasses = 2, missingLevel = 0; boolean predictorMissing = false, classMissing = false; Instances train1 = null; Instances train2 = null; ASSearch search = null; ASEvaluation evaluation1A = null; ASEvaluation evaluation1B = null; ASEvaluation evaluation2 = null; AttributeSelection attsel1A = null; AttributeSelection attsel1B = null; int stage = 0; try { // Make two train sets with different numbers of attributes train1 = makeTestDataset(42, numTrain, nominalPredictor ? getNumNominal() : 0, numericPredictor ? getNumNumeric() : 0, stringPredictor ? getNumString() : 0, datePredictor ? getNumDate() : 0, relationalPredictor ? getNumRelational() : 0, numClasses, classType, multiInstance); train2 = makeTestDataset(84, numTrain, nominalPredictor ? getNumNominal() + 1 : 0, numericPredictor ? getNumNumeric() + 1 : 0, stringPredictor ? getNumString() : 0, datePredictor ? getNumDate() : 0, relationalPredictor ? getNumRelational() : 0, numClasses, classType, multiInstance); if (missingLevel > 0) { addMissing(train1, missingLevel, predictorMissing, classMissing); addMissing(train2, missingLevel, predictorMissing, classMissing); } search = ASSearch.makeCopies(getSearch(), 1)[0]; evaluation1A = ASEvaluation.makeCopies(getEvaluator(), 1)[0]; evaluation1B = ASEvaluation.makeCopies(getEvaluator(), 1)[0]; evaluation2 = ASEvaluation.makeCopies(getEvaluator(), 1)[0]; } catch (Exception ex) { throw new Error("Error setting up for tests: " + ex.getMessage()); } try { stage = 0; attsel1A = search(search, evaluation1A, train1); stage = 1; search(search, evaluation2, train2); stage = 2; attsel1B = search(search, evaluation1B, train1); stage = 3; if (!attsel1A.toResultsString().equals(attsel1B.toResultsString())) { if (m_Debug) { println( "\n=== Full report ===\n" + "\nFirst search\n" + attsel1A.toResultsString() + "\n\n"); println( "\nSecond search\n" + attsel1B.toResultsString() + "\n\n"); } throw new Exception("Results differ between search calls"); } println("yes"); result[0] = true; if (false && m_Debug) { println( "\n=== Full report ===\n" + "\nFirst search\n" + evaluation1A.toString() + "\n\n"); println( "\nSecond search\n" + evaluation1B.toString() + "\n\n"); } } catch (Exception ex) { println("no"); result[0] = false; if (m_Debug) { println("\n=== Full Report ==="); print("Problem during training"); switch (stage) { case 0: print(" of dataset 1"); break; case 1: print(" of dataset 2"); break; case 2: print(" of dataset 1 (2nd build)"); break; case 3: print(", comparing results from builds of dataset 1"); break; } println(": " + ex.getMessage() + "\n"); println("here are the datasets:\n"); println("=== Train1 Dataset ===\n" + train1.toString() + "\n"); println("=== Train2 Dataset ===\n" + train2.toString() + "\n"); } } return result; } /** * Checks basic missing value handling of the scheme. If the missing * values cause an exception to be thrown by the scheme, this will be * recorded. * * @param nominalPredictor if true use nominal predictor attributes * @param numericPredictor if true use numeric predictor attributes * @param stringPredictor if true use string predictor attributes * @param datePredictor if true use date predictor attributes * @param relationalPredictor if true use relational predictor attributes * @param multiInstance whether multi-instance is needed * @param classType the class type (NUMERIC, NOMINAL, etc.) * @param predictorMissing true if the missing values may be in * the predictors * @param classMissing true if the missing values may be in the class * @param missingLevel the percentage of missing values * @return index 0 is true if the test was passed, index 1 is true if test * was acceptable */ protected boolean[] canHandleMissing( boolean nominalPredictor, boolean numericPredictor, boolean stringPredictor, boolean datePredictor, boolean relationalPredictor, boolean multiInstance, int classType, boolean predictorMissing, boolean classMissing, int missingLevel) { if (missingLevel == 100) print("100% "); print("missing"); if (predictorMissing) { print(" predictor"); if (classMissing) print(" and"); } if (classMissing) print(" class"); print(" values"); printAttributeSummary( nominalPredictor, numericPredictor, stringPredictor, datePredictor, relationalPredictor, multiInstance, classType); print("..."); FastVector accepts = new FastVector(); accepts.addElement("missing"); accepts.addElement("value"); accepts.addElement("train"); accepts.addElement("no attributes"); int numTrain = getNumInstances(), numClasses = 2; return runBasicTest(nominalPredictor, numericPredictor, stringPredictor, datePredictor, relationalPredictor, multiInstance, classType, missingLevel, predictorMissing, classMissing, numTrain, numClasses, accepts); } /** * Checks whether the scheme can handle instance weights. * This test compares the scheme performance on two datasets * that are identical except for the training weights. If the * results change, then the scheme must be using the weights. It * may be possible to get a false positive from this test if the * weight changes aren't significant enough to induce a change * in scheme performance (but the weights are chosen to minimize * the likelihood of this). * * @param nominalPredictor if true use nominal predictor attributes * @param numericPredictor if true use numeric predictor attributes * @param stringPredictor if true use string predictor attributes * @param datePredictor if true use date predictor attributes * @param relationalPredictor if true use relational predictor attributes * @param multiInstance whether multi-instance is needed * @param classType the class type (NUMERIC, NOMINAL, etc.) * @return index 0 true if the test was passed */ protected boolean[] instanceWeights( boolean nominalPredictor, boolean numericPredictor, boolean stringPredictor, boolean datePredictor, boolean relationalPredictor, boolean multiInstance, int classType) { print("scheme uses instance weights"); printAttributeSummary( nominalPredictor, numericPredictor, stringPredictor, datePredictor, relationalPredictor, multiInstance, classType); print("..."); int numTrain = 2*getNumInstances(), numClasses = 2, missingLevel = 0; boolean predictorMissing = false, classMissing = false; boolean[] result = new boolean[2]; Instances train = null; ASSearch[] search = null; ASEvaluation evaluationB = null; ASEvaluation evaluationI = null; AttributeSelection attselB = null; AttributeSelection attselI = null; boolean evalFail = false; try { train = makeTestDataset(42, numTrain, nominalPredictor ? getNumNominal() + 1 : 0, numericPredictor ? getNumNumeric() + 1 : 0, stringPredictor ? getNumString() : 0, datePredictor ? getNumDate() : 0, relationalPredictor ? getNumRelational() : 0, numClasses, classType, multiInstance); if (missingLevel > 0) addMissing(train, missingLevel, predictorMissing, classMissing); search = ASSearch.makeCopies(getSearch(), 2); evaluationB = ASEvaluation.makeCopies(getEvaluator(), 1)[0]; evaluationI = ASEvaluation.makeCopies(getEvaluator(), 1)[0]; attselB = search(search[0], evaluationB, train); } catch (Exception ex) { throw new Error("Error setting up for tests: " + ex.getMessage()); } try { // Now modify instance weights and re-built/test for (int i = 0; i < train.numInstances(); i++) { train.instance(i).setWeight(0); } Random random = new Random(1); for (int i = 0; i < train.numInstances() / 2; i++) { int inst = Math.abs(random.nextInt()) % train.numInstances(); int weight = Math.abs(random.nextInt()) % 10 + 1; train.instance(inst).setWeight(weight); } attselI = search(search[1], evaluationI, train); if (attselB.toResultsString().equals(attselI.toResultsString())) { // println("no"); evalFail = true; throw new Exception("evalFail"); } println("yes"); result[0] = true; } catch (Exception ex) { println("no"); result[0] = false; if (m_Debug) { println("\n=== Full Report ==="); if (evalFail) { println("Results don't differ between non-weighted and " + "weighted instance models."); println("Here are the results:\n"); println("\nboth methods\n"); println(evaluationB.toString()); } else { print("Problem during training"); println(": " + ex.getMessage() + "\n"); } println("Here is the dataset:\n"); println("=== Train Dataset ===\n" + train.toString() + "\n"); println("=== Train Weights ===\n"); for (int i = 0; i < train.numInstances(); i++) { println(" " + (i + 1) + " " + train.instance(i).weight()); } } } return result; } /** * Checks whether the scheme alters the training dataset during * training. If the scheme needs to modify the training * data it should take a copy of the training data. Currently checks * for changes to header structure, number of instances, order of * instances, instance weights. * * @param nominalPredictor if true use nominal predictor attributes * @param numericPredictor if true use numeric predictor attributes * @param stringPredictor if true use string predictor attributes * @param datePredictor if true use date predictor attributes * @param relationalPredictor if true use relational predictor attributes * @param multiInstance whether multi-instance is needed * @param classType the class type (NUMERIC, NOMINAL, etc.) * @param predictorMissing true if we know the scheme can handle * (at least) moderate missing predictor values * @param classMissing true if we know the scheme can handle * (at least) moderate missing class values * @return index 0 is true if the test was passed */ protected boolean[] datasetIntegrity( boolean nominalPredictor, boolean numericPredictor, boolean stringPredictor, boolean datePredictor, boolean relationalPredictor, boolean multiInstance, int classType, boolean predictorMissing, boolean classMissing) { print("scheme doesn't alter original datasets"); printAttributeSummary( nominalPredictor, numericPredictor, stringPredictor, datePredictor, relationalPredictor, multiInstance, classType); print("..."); int numTrain = getNumInstances(), numClasses = 2, missingLevel = 20; boolean[] result = new boolean[2]; Instances train = null; Instances trainCopy = null; ASSearch search = null; ASEvaluation evaluation = null; try { train = makeTestDataset(42, numTrain, nominalPredictor ? getNumNominal() : 0, numericPredictor ? getNumNumeric() : 0, stringPredictor ? getNumString() : 0, datePredictor ? getNumDate() : 0, relationalPredictor ? getNumRelational() : 0, numClasses, classType, multiInstance); if (missingLevel > 0) addMissing(train, missingLevel, predictorMissing, classMissing); search = ASSearch.makeCopies(getSearch(), 1)[0]; evaluation = ASEvaluation.makeCopies(getEvaluator(), 1)[0]; trainCopy = new Instances(train); } catch (Exception ex) { throw new Error("Error setting up for tests: " + ex.getMessage()); } try { search(search, evaluation, trainCopy); compareDatasets(train, trainCopy); println("yes"); result[0] = true; } catch (Exception ex) { println("no"); result[0] = false; if (m_Debug) { println("\n=== Full Report ==="); print("Problem during training"); println(": " + ex.getMessage() + "\n"); println("Here are the datasets:\n"); println("=== Train Dataset (original) ===\n" + trainCopy.toString() + "\n"); println("=== Train Dataset ===\n" + train.toString() + "\n"); } } return result; } /** * Runs a text on the datasets with the given characteristics. * * @param nominalPredictor if true use nominal predictor attributes * @param numericPredictor if true use numeric predictor attributes * @param stringPredictor if true use string predictor attributes * @param datePredictor if true use date predictor attributes * @param relationalPredictor if true use relational predictor attributes * @param multiInstance whether multi-instance is needed * @param classType the class type (NUMERIC, NOMINAL, etc.) * @param missingLevel the percentage of missing values * @param predictorMissing true if the missing values may be in * the predictors * @param classMissing true if the missing values may be in the class * @param numTrain the number of instances in the training set * @param numClasses the number of classes * @param accepts the acceptable string in an exception * @return index 0 is true if the test was passed, index 1 is true if test * was acceptable */ protected boolean[] runBasicTest(boolean nominalPredictor, boolean numericPredictor, boolean stringPredictor, boolean datePredictor, boolean relationalPredictor, boolean multiInstance, int classType, int missingLevel, boolean predictorMissing, boolean classMissing, int numTrain, int numClasses, FastVector accepts) { return runBasicTest( nominalPredictor, numericPredictor, stringPredictor, datePredictor, relationalPredictor, multiInstance, classType, TestInstances.CLASS_IS_LAST, missingLevel, predictorMissing, classMissing, numTrain, numClasses, accepts); } /** * Runs a text on the datasets with the given characteristics. * * @param nominalPredictor if true use nominal predictor attributes * @param numericPredictor if true use numeric predictor attributes * @param stringPredictor if true use string predictor attributes * @param datePredictor if true use date predictor attributes * @param relationalPredictor if true use relational predictor attributes * @param multiInstance whether multi-instance is needed * @param classType the class type (NUMERIC, NOMINAL, etc.) * @param classIndex the attribute index of the class * @param missingLevel the percentage of missing values * @param predictorMissing true if the missing values may be in * the predictors * @param classMissing true if the missing values may be in the class * @param numTrain the number of instances in the training set * @param numClasses the number of classes * @param accepts the acceptable string in an exception * @return index 0 is true if the test was passed, index 1 is true if test * was acceptable */ protected boolean[] runBasicTest(boolean nominalPredictor, boolean numericPredictor, boolean stringPredictor, boolean datePredictor, boolean relationalPredictor, boolean multiInstance, int classType, int classIndex, int missingLevel, boolean predictorMissing, boolean classMissing, int numTrain, int numClasses, FastVector accepts) { boolean[] result = new boolean[2]; Instances train = null; ASSearch search = null; ASEvaluation evaluation = null; try { train = makeTestDataset(42, numTrain, nominalPredictor ? getNumNominal() : 0, numericPredictor ? getNumNumeric() : 0, stringPredictor ? getNumString() : 0, datePredictor ? getNumDate() : 0, relationalPredictor ? getNumRelational() : 0, numClasses, classType, classIndex, multiInstance); if (missingLevel > 0) addMissing(train, missingLevel, predictorMissing, classMissing); search = ASSearch.makeCopies(getSearch(), 1)[0]; evaluation = ASEvaluation.makeCopies(getEvaluator(), 1)[0]; } catch (Exception ex) { ex.printStackTrace(); throw new Error("Error setting up for tests: " + ex.getMessage()); } try { search(search, evaluation, train); println("yes"); result[0] = true; } catch (Exception ex) { boolean acceptable = false; String msg; if (ex.getMessage() == null) msg = ""; else msg = ex.getMessage().toLowerCase(); if (msg.indexOf("not in classpath") > -1) m_ClasspathProblems = true; for (int i = 0; i < accepts.size(); i++) { if (msg.indexOf((String)accepts.elementAt(i)) >= 0) { acceptable = true; } } println("no" + (acceptable ? " (OK error message)" : "")); result[1] = acceptable; if (m_Debug) { println("\n=== Full Report ==="); print("Problem during training"); println(": " + ex.getMessage() + "\n"); if (!acceptable) { if (accepts.size() > 0) { print("Error message doesn't mention "); for (int i = 0; i < accepts.size(); i++) { if (i != 0) { print(" or "); } print('"' + (String)accepts.elementAt(i) + '"'); } } println("here is the dataset:\n"); println("=== Train Dataset ===\n" + train.toString() + "\n"); } } } return result; } /** * Make a simple set of instances, which can later be modified * for use in specific tests. * * @param seed the random number seed * @param numInstances the number of instances to generate * @param numNominal the number of nominal attributes * @param numNumeric the number of numeric attributes * @param numString the number of string attributes * @param numDate the number of date attributes * @param numRelational the number of relational attributes * @param numClasses the number of classes (if nominal class) * @param classType the class type (NUMERIC, NOMINAL, etc.) * @param multiInstance whether the dataset should a multi-instance dataset * @return the test dataset * @throws Exception if the dataset couldn't be generated * @see #process(Instances) */ protected Instances makeTestDataset(int seed, int numInstances, int numNominal, int numNumeric, int numString, int numDate, int numRelational, int numClasses, int classType, boolean multiInstance) throws Exception { return makeTestDataset( seed, numInstances, numNominal, numNumeric, numString, numDate, numRelational, numClasses, classType, TestInstances.CLASS_IS_LAST, multiInstance); } /** * Make a simple set of instances with variable position of the class * attribute, which can later be modified for use in specific tests. * * @param seed the random number seed * @param numInstances the number of instances to generate * @param numNominal the number of nominal attributes * @param numNumeric the number of numeric attributes * @param numString the number of string attributes * @param numDate the number of date attributes * @param numRelational the number of relational attributes * @param numClasses the number of classes (if nominal class) * @param classType the class type (NUMERIC, NOMINAL, etc.) * @param classIndex the index of the class (0-based, -1 as last) * @param multiInstance whether the dataset should a multi-instance dataset * @return the test dataset * @throws Exception if the dataset couldn't be generated * @see TestInstances#CLASS_IS_LAST * @see #process(Instances) */ protected Instances makeTestDataset(int seed, int numInstances, int numNominal, int numNumeric, int numString, int numDate, int numRelational, int numClasses, int classType, int classIndex, boolean multiInstance) throws Exception { TestInstances dataset = new TestInstances(); dataset.setSeed(seed); dataset.setNumInstances(numInstances); dataset.setNumNominal(numNominal); dataset.setNumNumeric(numNumeric); dataset.setNumString(numString); dataset.setNumDate(numDate); dataset.setNumRelational(numRelational); dataset.setNumClasses(numClasses); dataset.setClassType(classType); dataset.setClassIndex(classIndex); dataset.setNumClasses(numClasses); dataset.setMultiInstance(multiInstance); dataset.setWords(getWords()); dataset.setWordSeparators(getWordSeparators()); return process(dataset.generate()); } /** * Print out a short summary string for the dataset characteristics * * @param nominalPredictor true if nominal predictor attributes are present * @param numericPredictor true if numeric predictor attributes are present * @param stringPredictor true if string predictor attributes are present * @param datePredictor true if date predictor attributes are present * @param relationalPredictor true if relational predictor attributes are present * @param multiInstance whether multi-instance is needed * @param classType the class type (NUMERIC, NOMINAL, etc.) */ protected void printAttributeSummary(boolean nominalPredictor, boolean numericPredictor, boolean stringPredictor, boolean datePredictor, boolean relationalPredictor, boolean multiInstance, int classType) { String str = ""; if (numericPredictor) str += " numeric"; if (nominalPredictor) { if (str.length() > 0) str += " &"; str += " nominal"; } if (stringPredictor) { if (str.length() > 0) str += " &"; str += " string"; } if (datePredictor) { if (str.length() > 0) str += " &"; str += " date"; } if (relationalPredictor) { if (str.length() > 0) str += " &"; str += " relational"; } str += " predictors)"; switch (classType) { case Attribute.NUMERIC: str = " (numeric class," + str; break; case Attribute.NOMINAL: str = " (nominal class," + str; break; case Attribute.STRING: str = " (string class," + str; break; case Attribute.DATE: str = " (date class," + str; break; case Attribute.RELATIONAL: str = " (relational class," + str; break; } print(str); } /** * Returns the revision string. * * @return the revision */ public String getRevision() { return RevisionUtils.extract("$Revision: 4783 $"); } /** * Test method for this class * * @param args the commandline parameters */ public static void main(String [] args) { runCheck(new CheckAttributeSelection(), args); } }