yunfengzhou-hub commented on code in PR #187: URL: https://github.com/apache/flink-ml/pull/187#discussion_r1041657706
########## flink-ml-lib/src/main/java/org/apache/flink/ml/feature/univariatefeatureselector/UnivariateFeatureSelector.java: ########## @@ -0,0 +1,303 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.flink.ml.feature.univariatefeatureselector; + +import org.apache.flink.api.common.state.ListState; +import org.apache.flink.api.common.state.ListStateDescriptor; +import org.apache.flink.api.common.typeinfo.TypeInformation; +import org.apache.flink.api.common.typeinfo.Types; +import org.apache.flink.api.java.tuple.Tuple2; +import org.apache.flink.ml.api.Estimator; +import org.apache.flink.ml.param.Param; +import org.apache.flink.ml.stats.anovatest.ANOVATest; +import org.apache.flink.ml.stats.chisqtest.ChiSqTest; +import org.apache.flink.ml.stats.fvaluetest.FValueTest; +import org.apache.flink.ml.util.ParamUtils; +import org.apache.flink.ml.util.ReadWriteUtils; +import org.apache.flink.runtime.state.StateInitializationContext; +import org.apache.flink.runtime.state.StateSnapshotContext; +import org.apache.flink.streaming.api.datastream.DataStream; +import org.apache.flink.streaming.api.operators.AbstractStreamOperator; +import org.apache.flink.streaming.api.operators.BoundedOneInput; +import org.apache.flink.streaming.api.operators.OneInputStreamOperator; +import org.apache.flink.streaming.runtime.streamrecord.StreamRecord; +import org.apache.flink.table.api.Table; +import org.apache.flink.table.api.bridge.java.StreamTableEnvironment; +import org.apache.flink.table.api.internal.TableImpl; +import org.apache.flink.types.Row; +import org.apache.flink.util.Preconditions; + +import org.apache.commons.collections.IteratorUtils; + +import java.io.IOException; +import java.util.ArrayList; +import java.util.Comparator; +import java.util.HashMap; +import java.util.List; +import java.util.Map; +import java.util.stream.IntStream; + +/** + * An Estimator which selects features based on univariate statistical tests against labels. + * + * <p>Currently, Flink supports three Univariate Feature Selectors: chi-squared, ANOVA F-test and + * F-value. User can choose Univariate Feature Selector by setting `featureType` and `labelType`, + * and Flink will pick the score function based on the specified `featureType` and `labelType`. + * + * <p>The following combination of `featureType` and `labelType` are supported: + * + * <ul> + * <li>`featureType` `categorical` and `labelType` `categorical`: Flink uses chi-squared, i.e. + * chi2 in sklearn. + * <li>`featureType` `continuous` and `labelType` `categorical`: Flink uses ANOVA F-test, + * f_classif in sklearn. + * <li>`featureType` `continuous` and `labelType` `continuous`: Flink uses F-value, i.e. + * f_regression in sklearn. + * </ul> + * + * <p>The `UnivariateFeatureSelector` supports different selection modes: + * + * <ul> + * <li>numTopFeatures: chooses a fixed number of top features according to a hypothesis. + * <li>percentile: similar to numTopFeatures but chooses a fraction of all features instead of a + * fixed number. + * <li>fpr: chooses all features whose p-value are below a threshold, thus controlling the false + * positive rate of selection. + * <li>fdr: uses the <a + * href="https://en.wikipedia.org/wiki/False_discovery_rate#Benjamini.E2.80.93Hochberg_procedure";> + * Benjamini-Hochberg procedure</a> to choose all features whose false discovery rate is below + * a threshold. + * <li>fwe: chooses all features whose p-values are below a threshold. The threshold is scaled by + * 1/numFeatures, thus controlling the family-wise error rate of selection. + * </ul> + * + * <p>By default, the selection mode is `numTopFeatures`. + */ +public class UnivariateFeatureSelector + implements Estimator<UnivariateFeatureSelector, UnivariateFeatureSelectorModel>, + UnivariateFeatureSelectorParams<UnivariateFeatureSelector> { + + private final Map<Param<?>, Object> paramMap = new HashMap<>(); + + public UnivariateFeatureSelector() { + ParamUtils.initializeMapWithDefaultValues(paramMap, this); + } + + @Override + public UnivariateFeatureSelectorModel fit(Table... inputs) { + Preconditions.checkArgument(inputs.length == 1); + + final String featuresCol = getFeaturesCol(); + final String labelCol = getLabelCol(); + final String featureType = getFeatureType(); + final String labelType = getLabelType(); + + StreamTableEnvironment tEnv = + (StreamTableEnvironment) ((TableImpl) inputs[0]).getTableEnvironment(); + + Table output; + if (CATEGORICAL.equals(featureType) && CATEGORICAL.equals(labelType)) { + output = + new ChiSqTest() + .setFeaturesCol(featuresCol) + .setLabelCol(labelCol) + .setFlatten(true) + .transform(inputs[0])[0]; + } else if (CONTINUOUS.equals(featureType) && CATEGORICAL.equals(labelType)) { + output = + new ANOVATest() + .setFeaturesCol(featuresCol) + .setLabelCol(labelCol) + .setFlatten(true) + .transform(inputs[0])[0]; + } else if (CONTINUOUS.equals(featureType) && CONTINUOUS.equals(labelType)) { + output = + new FValueTest() + .setFeaturesCol(featuresCol) + .setLabelCol(labelCol) + .setFlatten(true) + .transform(inputs[0])[0]; + } else { + throw new IllegalArgumentException( + String.format( + "Unsupported combination: featureType=%s, labelType=%s.", + featureType, labelType)); + } + DataStream<UnivariateFeatureSelectorModelData> modelData = + tEnv.toDataStream(output) + .transform( + "selectIndicesFromPValues", + TypeInformation.of(UnivariateFeatureSelectorModelData.class), + new SelectIndicesFromPValuesOperator( + getSelectionMode(), getActualSelectionThreshold())) + .setParallelism(1); + UnivariateFeatureSelectorModel model = + new UnivariateFeatureSelectorModel().setModelData(tEnv.fromDataStream(modelData)); + ReadWriteUtils.updateExistingParams(model, getParamMap()); + return model; + } + + private double getActualSelectionThreshold() { + Double threshold = getSelectionThreshold(); + if (threshold == null) { + String selectionMode = getSelectionMode(); + if (NUM_TOP_FEATURES.equals(selectionMode)) { + threshold = 50.0; + } else if (PERCENTILE.equals(selectionMode)) { + threshold = 0.1; + } else { + threshold = 0.05; + } + } else { + if (NUM_TOP_FEATURES.equals(getSelectionMode())) { + Preconditions.checkArgument( + threshold >= 1 && threshold.intValue() == threshold, + "SelectionThreshold needs to be a positive Integer " + + "for selection mode numTopFeatures, but got %s.", + threshold); + } else { + Preconditions.checkArgument( + threshold >= 0 && threshold <= 1, + "SelectionThreshold needs to be in the range [0, 1] " + + "for selection mode %s, but got %s.", + getSelectionMode(), + threshold); + } + } + return threshold; + } + + private static class SelectIndicesFromPValuesOperator + extends AbstractStreamOperator<UnivariateFeatureSelectorModelData> + implements OneInputStreamOperator<Row, UnivariateFeatureSelectorModelData>, + BoundedOneInput { + private final String selectionMode; + private final double threshold; + + private List<Tuple2<Double, Integer>> pValuesAndIndices; + private ListState<Tuple2<Double, Integer>> pValuesAndIndicesState; + + public SelectIndicesFromPValuesOperator(String selectionMode, double threshold) { + this.selectionMode = selectionMode; + this.threshold = threshold; + } + + @Override + public void endInput() { + List<Integer> indices = new ArrayList<>(); + + switch (selectionMode) { + case NUM_TOP_FEATURES: + pValuesAndIndices.sort( + Comparator.comparingDouble((Tuple2<Double, Integer> t) -> t.f0) + .thenComparingInt(t -> t.f1)); + IntStream.range(0, Math.min(pValuesAndIndices.size(), (int) threshold)) + .forEach(i -> indices.add(pValuesAndIndices.get(i).f1)); + break; + case PERCENTILE: + pValuesAndIndices.sort( + Comparator.comparingDouble((Tuple2<Double, Integer> t) -> t.f0) + .thenComparingInt(t -> t.f1)); + IntStream.range( + 0, + Math.min( + pValuesAndIndices.size(), + (int) (pValuesAndIndices.size() * threshold))) + .forEach(i -> indices.add(pValuesAndIndices.get(i).f1)); + break; + case FPR: + pValuesAndIndices.stream() + .filter(x -> x.f0 < threshold) + .forEach(x -> indices.add(x.f1)); + break; + case FDR: + pValuesAndIndices.sort( + Comparator.comparingDouble((Tuple2<Double, Integer> t) -> t.f0) + .thenComparingInt(t -> t.f1)); + + int maxIndex = -1; + for (int i = 0; i < pValuesAndIndices.size(); i++) { + if (pValuesAndIndices.get(i).f0 + < (threshold / pValuesAndIndices.size()) * (i + 1)) { + maxIndex = Math.max(maxIndex, i); + } + } + if (maxIndex >= 0) { + pValuesAndIndices.sort(Comparator.comparing(t -> t.f0)); Review Comment: Let's compare pValues and then indices in FDR strategy as well. ########## flink-ml-lib/src/main/java/org/apache/flink/ml/feature/variancethresholdselector/VarianceThresholdSelector.java: ########## @@ -63,11 +63,11 @@ public VarianceThresholdSelectorModel fit(Table... inputs) { final String inputCol = getInputCol(); StreamTableEnvironment tEnv = (StreamTableEnvironment) ((TableImpl) inputs[0]).getTableEnvironment(); - DataStream<DenseVector> inputData = + DataStream<Vector> inputData = Review Comment: Let's describe the changes made to `VarianceThresholdSelector` in this PR's description. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
