zhipeng93 commented on code in PR #237: URL: https://github.com/apache/flink-ml/pull/237#discussion_r1209952491
########## flink-ml-lib/src/main/java/org/apache/flink/ml/classification/logisticregression/LogisticRegressionWithFtrl.java: ########## @@ -0,0 +1,380 @@ +/* + * 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.classification.logisticregression; + +import org.apache.flink.api.common.functions.MapFunction; +import org.apache.flink.api.common.functions.ReduceFunction; +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.java.tuple.Tuple2; +import org.apache.flink.api.java.tuple.Tuple3; +import org.apache.flink.ml.api.Estimator; +import org.apache.flink.ml.common.datastream.DataStreamUtils; +import org.apache.flink.ml.common.feature.LabeledLargePointWithWeight; +import org.apache.flink.ml.common.lossfunc.BinaryLogisticLoss; +import org.apache.flink.ml.common.lossfunc.LossFunc; +import org.apache.flink.ml.common.ps.training.IterationStageList; +import org.apache.flink.ml.common.ps.training.ProcessStage; +import org.apache.flink.ml.common.ps.training.PullStage; +import org.apache.flink.ml.common.ps.training.PushStage; +import org.apache.flink.ml.common.ps.training.SerializableConsumer; +import org.apache.flink.ml.common.ps.training.TrainingContext; +import org.apache.flink.ml.common.ps.training.TrainingUtils; +import org.apache.flink.ml.common.updater.FTRL; +import org.apache.flink.ml.linalg.Vectors; +import org.apache.flink.ml.param.Param; +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.runtime.util.ResettableIterator; +import org.apache.flink.streaming.api.datastream.DataStream; +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.flink.util.function.SerializableFunction; +import org.apache.flink.util.function.SerializableSupplier; + +import it.unimi.dsi.fastutil.longs.Long2DoubleOpenHashMap; +import it.unimi.dsi.fastutil.longs.LongOpenHashSet; +import org.apache.commons.collections.IteratorUtils; + +import java.io.IOException; +import java.util.ArrayList; +import java.util.Arrays; +import java.util.HashMap; +import java.util.Iterator; +import java.util.List; +import java.util.Map; + +/** + * An Estimator which implements the large scale logistic regression algorithm using FTRL optimizer. + * + * <p>See https://en.wikipedia.org/wiki/Logistic_regression. + */ +public class LogisticRegressionWithFtrl + implements Estimator<LogisticRegressionWithFtrl, LogisticRegressionModel>, + LogisticRegressionWithFtrlParams<LogisticRegressionWithFtrl> { + + private final Map<Param<?>, Object> paramMap = new HashMap<>(); + + public LogisticRegressionWithFtrl() { + ParamUtils.initializeMapWithDefaultValues(paramMap, this); + } + + @Override + public LogisticRegressionModel fit(Table... inputs) { + Preconditions.checkArgument(inputs.length == 1); + String classificationType = getMultiClass(); + Preconditions.checkArgument( + "auto".equals(classificationType) || "binomial".equals(classificationType), + "Multinomial classification is not supported yet. Supported options: [auto, binomial]."); + StreamTableEnvironment tEnv = + (StreamTableEnvironment) ((TableImpl) inputs[0]).getTableEnvironment(); + + DataStream<LabeledLargePointWithWeight> trainData = + tEnv.toDataStream(inputs[0]) + .map( + (MapFunction<Row, LabeledLargePointWithWeight>) + dataPoint -> { + double weight = + getWeightCol() == null + ? 1.0 + : ((Number) + dataPoint.getField( + getWeightCol())) + .doubleValue(); + double label = + ((Number) dataPoint.getField(getLabelCol())) + .doubleValue(); + boolean isBinomial = + Double.compare(0., label) == 0 + || Double.compare(1., label) == 0; + if (!isBinomial) { + throw new RuntimeException( + "Multinomial classification is not supported yet. Supported options: [auto, binomial]."); + } + Tuple2<long[], double[]> features = + dataPoint.getFieldAs(getFeaturesCol()); + return new LabeledLargePointWithWeight( + features, label, weight); + }); + + DataStream<Long> modelDim; + if (getModelDim() > 0) { + modelDim = trainData.getExecutionEnvironment().fromElements(getModelDim()); + } else { + modelDim = + DataStreamUtils.reduce( + trainData.map(x -> x.features.f0[x.features.f0.length - 1]), + (ReduceFunction<Long>) Math::max) + .map((MapFunction<Long, Long>) value -> value + 1); + } + + LogisticRegressionWithFtrlTrainingContext trainingContext = + new LogisticRegressionWithFtrlTrainingContext(getParamMap()); + + IterationStageList<LogisticRegressionWithFtrlTrainingContext> iterationStages = + new IterationStageList<>(trainingContext); + iterationStages + .addTrainingStage(new ComputeIndices()) + .addTrainingStage( + new PullStage( + (SerializableSupplier<long[]>) () -> trainingContext.pullIndices, + (SerializableConsumer<double[]>) + x -> trainingContext.pulledValues = x)) + .addTrainingStage(new ComputeGradients(BinaryLogisticLoss.INSTANCE)) + .addTrainingStage( + new PushStage( + (SerializableSupplier<long[]>) () -> trainingContext.pushIndices, + (SerializableSupplier<double[]>) () -> trainingContext.pushValues)) + .setTerminationCriteria( + (SerializableFunction<LogisticRegressionWithFtrlTrainingContext, Boolean>) + o -> o.iterationId >= getMaxIter()); + FTRL ftrl = new FTRL(getAlpha(), getBeta(), getReg(), getElasticNet()); + + DataStream<Tuple3<Long, Long, double[]>> rawModelData = + TrainingUtils.train( + modelDim, + trainData, + ftrl, + iterationStages, + getNumServers(), + getNumServerCores()); + + final long modelVersion = 0L; + + DataStream<LogisticRegressionModelData> modelData = + rawModelData.map( + tuple3 -> + new LogisticRegressionModelData( + Vectors.dense(tuple3.f2), + tuple3.f0, + tuple3.f1, + modelVersion)); + + LogisticRegressionModel model = + new LogisticRegressionModel().setModelData(tEnv.fromDataStream(modelData)); + ParamUtils.updateExistingParams(model, paramMap); + return model; + } + + @Override + public void save(String path) throws IOException { + ReadWriteUtils.saveMetadata(this, path); + } + + public static LogisticRegressionWithFtrl load(StreamTableEnvironment tEnv, String path) + throws IOException { + return ReadWriteUtils.loadStageParam(path); + } + + @Override + public Map<Param<?>, Object> getParamMap() { + return paramMap; + } +} + +/** + * An iteration stage that samples a batch of training data and computes the indices needed to + * compute gradients. + */ +class ComputeIndices extends ProcessStage<LogisticRegressionWithFtrlTrainingContext> { + + @Override + public void process(LogisticRegressionWithFtrlTrainingContext context) throws Exception { + context.readInNextBatchData(); + context.pullIndices = computeIndices(context.batchData); + } + + public static long[] computeIndices(List<LabeledLargePointWithWeight> dataPoints) { + LongOpenHashSet indices = new LongOpenHashSet(); + for (LabeledLargePointWithWeight dataPoint : dataPoints) { + long[] notZeros = dataPoint.features.f0; + for (long index : notZeros) { + indices.add(index); + } + } + + long[] sortedIndices = new long[indices.size()]; + Iterator<Long> iterator = indices.iterator(); + int i = 0; + while (iterator.hasNext()) { + sortedIndices[i++] = iterator.next(); + } + Arrays.sort(sortedIndices); + return sortedIndices; + } +} + +/** + * An iteration stage that uses the pulled model values and sampled batch data to compute the + * gradients. + */ +class ComputeGradients extends ProcessStage<LogisticRegressionWithFtrlTrainingContext> { Review Comment: We can move these two classes outside `LogisticRegressionWithFtrl`. The problem is that `LabeledLargePointWithWeight` is universally applicable. I have moved these two classes to `org.apache.flink.ml.common.ps.training` for now. We can get back to this discussion when we achieved consensus on `LabeledLargePointWithWeight`. -- 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]
