Examples with RandomData org.apache.commons.math.random.RandomData used on opensource projects

Example 21 with RandomData

use of org.apache.commons.math.random.RandomData in project knime-core by knime.

the class TreeNumericColumnDataTest method testCalcBestSplitClassificationSplitAtStart.

/**
 * Test splits at last possible split position - even if no change in target can be observed, see example data in
 * method body.
 * @throws Exception
 */
@Test
public void testCalcBestSplitClassificationSplitAtStart() throws Exception {
    // Index:  1 2 3 4 5 6 7
    // Value:  1 1 1|2 2|3 3
    // Target: A A A|A A|A B
    double[] data = asDataArray("1,1,1,2,2,3,3");
    String[] target = asStringArray("A,A,A,A,B,A,B");
    TreeEnsembleLearnerConfiguration config = createConfig();
    Pair<TreeOrdinaryNumericColumnData, TreeTargetNominalColumnData> exampleData = exampleData(config, data, target);
    TreeNumericColumnData columnData = exampleData.getFirst();
    TreeTargetNominalColumnData targetData = exampleData.getSecond();
    double[] rowWeights = new double[data.length];
    Arrays.fill(rowWeights, 1.0);
    TreeData treeData = createTreeDataClassification(exampleData);
    IDataIndexManager indexManager = new DefaultDataIndexManager(treeData);
    DataMemberships dataMemberships = new RootDataMemberships(rowWeights, treeData, indexManager);
    ClassificationPriors priors = targetData.getDistribution(rowWeights, config);
    RandomData rd = config.createRandomData();
    SplitCandidate splitCandidate = columnData.calcBestSplitClassification(dataMemberships, priors, targetData, rd);
    double gain = (1.0 - Math.pow(5.0 / 7.0, 2.0) - Math.pow(2.0 / 7.0, 2.0)) - 0.0 - 4.0 / 7.0 * (1.0 - Math.pow(2.0 / 4.0, 2.0) - Math.pow(2.0 / 4.0, 2.0));
    // manually calculated
    assertEquals(gain, splitCandidate.getGainValue(), 0.000001);
    NumericSplitCandidate numSplitCandidate = (NumericSplitCandidate) splitCandidate;
    TreeNodeNumericCondition[] childConditions = numSplitCandidate.getChildConditions();
    assertEquals(2, childConditions.length);
    assertEquals((1.0 + 2.0) / 2.0, childConditions[0].getSplitValue(), 0.0);
}

Example 22 with RandomData

use of org.apache.commons.math.random.RandomData in project knime-core by knime.

the class TreeNumericColumnDataTest method testXGBoostMissingValueHandling.

/**
 * This method tests if the conditions for child nodes are correct in case of XGBoostMissingValueHandling
 *
 * @throws Exception
 */
@Test
public void testXGBoostMissingValueHandling() throws Exception {
    TreeEnsembleLearnerConfiguration config = createConfig();
    config.setMissingValueHandling(MissingValueHandling.XGBoost);
    final TestDataGenerator dataGen = new TestDataGenerator(config);
    final RandomData rd = config.createRandomData();
    final int[] indices = new int[] { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 };
    final double[] weights = new double[10];
    Arrays.fill(weights, 1.0);
    final MockDataColMem dataMem = new MockDataColMem(indices, indices, weights);
    final String dataCSV = "1,2,2,3,4,5,6,7,NaN,NaN";
    final String target1CSV = "A,A,A,A,B,B,B,B,A,A";
    final String target2CSV = "A,A,A,A,B,B,B,B,B,B";
    final double expectedGain = 0.48;
    final TreeNumericColumnData col = dataGen.createNumericAttributeColumn(dataCSV, "testCol", 0);
    final TreeTargetNominalColumnData target1 = TestDataGenerator.createNominalTargetColumn(target1CSV);
    final SplitCandidate split1 = col.calcBestSplitClassification(dataMem, target1.getDistribution(weights, config), target1, rd);
    assertEquals("Wrong gain.", expectedGain, split1.getGainValue(), 1e-8);
    final TreeNodeCondition[] childConds1 = split1.getChildConditions();
    final TreeNodeNumericCondition numCondLeft1 = (TreeNodeNumericCondition) childConds1[0];
    assertEquals("Wrong split point.", 3.5, numCondLeft1.getSplitValue(), 1e-8);
    assertTrue("Missings were not sent in the correct direction.", numCondLeft1.acceptsMissings());
    final TreeNodeNumericCondition numCondRight1 = (TreeNodeNumericCondition) childConds1[1];
    assertEquals("Wrong split point.", 3.5, numCondRight1.getSplitValue(), 1e-8);
    assertFalse("Missings were not sent in the correct direction.", numCondRight1.acceptsMissings());
    final TreeTargetNominalColumnData target2 = TestDataGenerator.createNominalTargetColumn(target2CSV);
    final SplitCandidate split2 = col.calcBestSplitClassification(dataMem, target2.getDistribution(weights, config), target2, rd);
    assertEquals("Wrong gain.", expectedGain, split2.getGainValue(), 1e-8);
    final TreeNodeCondition[] childConds2 = split2.getChildConditions();
    final TreeNodeNumericCondition numCondLeft2 = (TreeNodeNumericCondition) childConds2[0];
    assertEquals("Wrong split point.", 3.5, numCondLeft2.getSplitValue(), 1e-8);
    assertFalse("Missings were not sent in the correct direction.", numCondLeft2.acceptsMissings());
    final TreeNodeNumericCondition numCondRight2 = (TreeNodeNumericCondition) childConds2[1];
    assertEquals("Wrong split point.", 3.5, numCondRight2.getSplitValue(), 1e-8);
    assertTrue("Missings were not sent in the correct direction.", numCondRight2.acceptsMissings());
}

Example 23 with RandomData

use of org.apache.commons.math.random.RandomData in project knime-core by knime.

the class TreeNumericColumnDataTest method testCalcBestSplitRegression.

@Test
public void testCalcBestSplitRegression() throws InvalidSettingsException {
    String dataCSV = "1,2,3,4,5,6,7,8,9,10";
    String targetCSV = "1,5,4,4.3,6.5,6.5,4,3,3,4";
    TreeEnsembleLearnerConfiguration config = new TreeEnsembleLearnerConfiguration(true);
    config.setNrModels(1);
    config.setDataSelectionWithReplacement(false);
    config.setUseDifferentAttributesAtEachNode(false);
    config.setDataFractionPerTree(1.0);
    config.setColumnSamplingMode(ColumnSamplingMode.None);
    TestDataGenerator dataGen = new TestDataGenerator(config);
    RandomData rd = config.createRandomData();
    TreeTargetNumericColumnData target = TestDataGenerator.createNumericTargetColumn(targetCSV);
    TreeNumericColumnData attribute = dataGen.createNumericAttributeColumn(dataCSV, "test-col", 0);
    TreeData data = new TreeData(new TreeAttributeColumnData[] { attribute }, target, TreeType.Ordinary);
    double[] weights = new double[10];
    Arrays.fill(weights, 1.0);
    DataMemberships rootMem = new RootDataMemberships(weights, data, new DefaultDataIndexManager(data));
    SplitCandidate firstSplit = attribute.calcBestSplitRegression(rootMem, target.getPriors(rootMem, config), target, rd);
    // calculated via OpenOffice calc
    assertEquals(10.885444, firstSplit.getGainValue(), 1e-5);
    TreeNodeCondition[] firstConditions = firstSplit.getChildConditions();
    assertEquals(2, firstConditions.length);
    for (int i = 0; i < firstConditions.length; i++) {
        assertThat(firstConditions[i], instanceOf(TreeNodeNumericCondition.class));
        TreeNodeNumericCondition numCond = (TreeNodeNumericCondition) firstConditions[i];
        assertEquals(1.5, numCond.getSplitValue(), 0);
    }
    // left child contains only one row therefore only look at right child
    BitSet expectedInChild = new BitSet(10);
    expectedInChild.set(1, 10);
    BitSet inChild = attribute.updateChildMemberships(firstConditions[1], rootMem);
    assertEquals(expectedInChild, inChild);
    DataMemberships childMem = rootMem.createChildMemberships(inChild);
    SplitCandidate secondSplit = attribute.calcBestSplitRegression(childMem, target.getPriors(childMem, config), target, rd);
    assertEquals(6.883555, secondSplit.getGainValue(), 1e-5);
    TreeNodeCondition[] secondConditions = secondSplit.getChildConditions();
    for (int i = 0; i < secondConditions.length; i++) {
        assertThat(secondConditions[i], instanceOf(TreeNodeNumericCondition.class));
        TreeNodeNumericCondition numCond = (TreeNodeNumericCondition) secondConditions[i];
        assertEquals(6.5, numCond.getSplitValue(), 0);
    }
}

Example 24 with RandomData

use of org.apache.commons.math.random.RandomData in project knime-core by knime.

the class RFSubsetColumnSampleStrategy method getColumnSampleForTreeNode.

/**
 * {@inheritDoc}
 */
@Override
public ColumnSample getColumnSampleForTreeNode(final TreeNodeSignature treeNodeSignature) {
    short[] signature = treeNodeSignature.getSignaturePath();
    JDKRandomGenerator generator = new JDKRandomGenerator();
    generator.setSeed(m_seed);
    int[] newSeed = new int[signature.length];
    for (int i = 0; i < signature.length; i++) {
        for (int p = 0; p <= signature[i]; p++) {
            newSeed[i] = generator.nextInt();
        }
    }
    generator.setSeed(newSeed);
    int totalColCount = m_data.getColumns().length;
    RandomData rd = new RandomDataImpl(generator);
    int[] includes = rd.nextPermutation(totalColCount, m_subsetSize);
    Arrays.sort(includes);
    return new SubsetColumnSample(m_data, includes);
}

Example 25 with RandomData

use of org.apache.commons.math.random.RandomData in project knime-core by knime.

the class TreeNominalColumnDataTest method testCalcBestSplitClassificationBinaryPCAXGBoostMissingValueHandling.

/**
 * Tests the XGBoost missing value handling in the case of binary splits calculated with the pca method (multiple classes)
 *
 * @throws Exception
 */
@Test
public void testCalcBestSplitClassificationBinaryPCAXGBoostMissingValueHandling() throws Exception {
    final TreeEnsembleLearnerConfiguration config = createConfig(false);
    config.setMissingValueHandling(MissingValueHandling.XGBoost);
    final TestDataGenerator dataGen = new TestDataGenerator(config);
    final RandomData rd = config.createRandomData();
    // test the case that there are no missing values in the training data
    final String noMissingCSV = "a, a, a, b, b, b, b, c, c";
    final String noMissingTarget = "A, B, B, C, C, C, B, A, B";
    TreeNominalColumnData dataCol = dataGen.createNominalAttributeColumn(noMissingCSV, "noMissings", 0);
    TreeTargetNominalColumnData targetCol = TestDataGenerator.createNominalTargetColumn(noMissingTarget);
    DataMemberships dataMem = createMockDataMemberships(targetCol.getNrRows());
    SplitCandidate split = dataCol.calcBestSplitClassification(dataMem, targetCol.getDistribution(dataMem, config), targetCol, rd);
    assertNotNull("There is a possible split.", split);
    assertEquals("Incorrect gain.", 0.2086, split.getGainValue(), 1e-3);
    assertThat(split, instanceOf(NominalBinarySplitCandidate.class));
    NominalBinarySplitCandidate nomSplit = (NominalBinarySplitCandidate) split;
    assertTrue("No missing values in the column.", nomSplit.getMissedRows().isEmpty());
    TreeNodeNominalBinaryCondition[] conditions = nomSplit.getChildConditions();
    assertEquals("A binary split must have 2 child conditions.", 2, conditions.length);
    String[] values = new String[] { "a", "c" };
    assertArrayEquals("Wrong values in child condition.", values, conditions[0].getValues());
    assertArrayEquals("Wrong values in child condition.", values, conditions[1].getValues());
    assertEquals("Wrong set logic.", SetLogic.IS_NOT_IN, conditions[0].getSetLogic());
    assertEquals("Wrong set logic.", SetLogic.IS_IN, conditions[1].getSetLogic());
    assertFalse("Missing values should be sent to the majority child (i.e. right)", conditions[0].acceptsMissings());
    assertTrue("Missing values should be sent to the majority child (i.e. right)", conditions[1].acceptsMissings());
    // test the case that there are missing values in the training data
    final String missingCSV = "a, a, a, b, b, b, b, c, c, ?";
    final String missingTarget = "A, B, B, C, C, C, B, A, B, C";
    dataCol = dataGen.createNominalAttributeColumn(missingCSV, "missings", 0);
    targetCol = TestDataGenerator.createNominalTargetColumn(missingTarget);
    dataMem = createMockDataMemberships(targetCol.getNrRows());
    split = dataCol.calcBestSplitClassification(dataMem, targetCol.getDistribution(dataMem, config), targetCol, rd);
    assertNotNull("There is a possible split.", split);
    assertEquals("Incorrect gain.", 0.24, split.getGainValue(), 1e-3);
    assertThat(split, instanceOf(NominalBinarySplitCandidate.class));
    nomSplit = (NominalBinarySplitCandidate) split;
    assertTrue("Split should handle missing values.", nomSplit.getMissedRows().isEmpty());
    conditions = nomSplit.getChildConditions();
    assertEquals("Wrong number of child conditions.", 2, conditions.length);
    assertArrayEquals("Wrong values in child condition.", values, conditions[0].getValues());
    assertArrayEquals("Wrong values in child condition.", values, conditions[1].getValues());
    assertEquals("Wrong set logic.", SetLogic.IS_NOT_IN, conditions[0].getSetLogic());
    assertEquals("Wrong set logic.", SetLogic.IS_IN, conditions[1].getSetLogic());
    assertTrue("Missing values should be sent to left child", conditions[0].acceptsMissings());
    assertFalse("Missing values should be sent to left child", conditions[1].acceptsMissings());
}