lindong28 commented on a change in pull request #28: URL: https://github.com/apache/flink-ml/pull/28#discussion_r758920274
########## File path: flink-ml-lib/src/main/java/org/apache/flink/ml/classification/linear/LogisticRegression.java ########## @@ -0,0 +1,653 @@ +/* + * 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.linear; + +import org.apache.flink.api.common.functions.RichMapFunction; +import org.apache.flink.api.common.state.ListState; +import org.apache.flink.api.common.state.ListStateDescriptor; +import org.apache.flink.api.common.typeinfo.BasicTypeInfo; +import org.apache.flink.api.common.typeinfo.PrimitiveArrayTypeInfo; +import org.apache.flink.api.common.typeinfo.TypeInformation; +import org.apache.flink.api.common.typeutils.base.DoubleComparator; +import org.apache.flink.api.java.tuple.Tuple2; +import org.apache.flink.api.java.tuple.Tuple3; +import org.apache.flink.api.java.typeutils.TupleTypeInfo; +import org.apache.flink.iteration.DataStreamList; +import org.apache.flink.iteration.IterationBody; +import org.apache.flink.iteration.IterationBodyResult; +import org.apache.flink.iteration.IterationConfig; +import org.apache.flink.iteration.IterationConfig.OperatorLifeCycle; +import org.apache.flink.iteration.IterationListener; +import org.apache.flink.iteration.Iterations; +import org.apache.flink.iteration.ReplayableDataStreamList; +import org.apache.flink.iteration.operator.OperatorStateUtils; +import org.apache.flink.ml.api.Estimator; +import org.apache.flink.ml.common.broadcast.BroadcastUtils; +import org.apache.flink.ml.common.datastream.DataStreamUtils; +import org.apache.flink.ml.common.iteration.TerminateOnMaxIterOrTol; +import org.apache.flink.ml.linalg.BLAS; +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.streaming.api.datastream.DataStream; +import org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator; +import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment; +import org.apache.flink.streaming.api.operators.AbstractStreamOperator; +import org.apache.flink.streaming.api.operators.AbstractUdfStreamOperator; +import org.apache.flink.streaming.api.operators.BoundedMultiInput; +import org.apache.flink.streaming.api.operators.BoundedOneInput; +import org.apache.flink.streaming.api.operators.OneInputStreamOperator; +import org.apache.flink.streaming.api.operators.TwoInputStreamOperator; +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.util.Collector; +import org.apache.flink.util.OutputTag; +import org.apache.flink.util.Preconditions; + +import org.apache.commons.collections.IteratorUtils; + +import java.io.IOException; +import java.util.ArrayList; +import java.util.Arrays; +import java.util.Collections; +import java.util.HashMap; +import java.util.List; +import java.util.Map; +import java.util.Random; + +/** + * This class implements methods to train a logistic regression model. For details, see + * https://en.wikipedia.org/wiki/Logistic_regression. + */ +public class LogisticRegression + implements Estimator<LogisticRegression, LogisticRegressionModel>, + LogisticRegressionParams<LogisticRegression> { + + private Map<Param<?>, Object> paramMap = new HashMap<>(); + + private static final OutputTag<Tuple2<double[], double[]>> MODEL_OUTPUT = + new OutputTag<Tuple2<double[], double[]>>("MODEL_OUTPUT") {}; + + public LogisticRegression() { + ParamUtils.initializeMapWithDefaultValues(this.paramMap, this); + } + + @Override + public Map<Param<?>, Object> getParamMap() { + return paramMap; + } + + @Override + public void save(String path) throws IOException { + ReadWriteUtils.saveMetadata(this, path); + } + + public static LogisticRegression load(StreamExecutionEnvironment env, String path) + throws IOException { + return ReadWriteUtils.loadStageParam(path); + } + + @Override + @SuppressWarnings("unchecked") + public LogisticRegressionModel fit(Table... inputs) { + Preconditions.checkArgument(inputs.length == 1); + StreamTableEnvironment tEnv = + (StreamTableEnvironment) ((TableImpl) inputs[0]).getTableEnvironment(); + + DataStream<Tuple3<Double, Double, double[]>> trainData = + tEnv.toDataStream(inputs[0]) + .map( + dataPoint -> + Tuple3.of( + getWeightCol() == null + ? new Double(1.0) + : (Double) + dataPoint.getField(getWeightCol()), + (Double) dataPoint.getField(getLabelCol()), + (double[]) dataPoint.getField(getFeaturesCol()))) + .returns( + new TupleTypeInfo<>( + BasicTypeInfo.DOUBLE_TYPE_INFO, + BasicTypeInfo.DOUBLE_TYPE_INFO, + PrimitiveArrayTypeInfo.DOUBLE_PRIMITIVE_ARRAY_TYPE_INFO)); + + DataStream<Double> distinctLabelValues = + DataStreamUtils.sortPartition( + DataStreamUtils.distinct(trainData.map(dataPoint -> dataPoint.f1)), + new DoubleComparator(true)); + final String broadcastLabelsName = "broadcastLabels"; + trainData = + BroadcastUtils.withBroadcastStream( + Collections.singletonList(trainData), + Collections.singletonMap(broadcastLabelsName, distinctLabelValues), + inputList -> { + DataStream data = inputList.get(0); + return data.transform( + "preProcess", + new TupleTypeInfo<>( + BasicTypeInfo.DOUBLE_TYPE_INFO, + BasicTypeInfo.DOUBLE_TYPE_INFO, + PrimitiveArrayTypeInfo + .DOUBLE_PRIMITIVE_ARRAY_TYPE_INFO), + new PreprocessDataOp( + new PreprocessOneRecord(broadcastLabelsName))); + }); + + DataStream<double[]> initModel = + trainData + .transform( + "genInitModel", + PrimitiveArrayTypeInfo.DOUBLE_PRIMITIVE_ARRAY_TYPE_INFO, + new GenInitModel()) + .returns(PrimitiveArrayTypeInfo.DOUBLE_PRIMITIVE_ARRAY_TYPE_INFO); + + DataStream<Tuple2<double[], double[]>> modelAndLoss = train(trainData, initModel); + + DataStream<LogisticRegressionModelData> modelData = + modelAndLoss + .connect(distinctLabelValues) + .transform( + "composeModelData", + TypeInformation.of(LogisticRegressionModelData.class), + new ComposeModelDataOp()) + .setParallelism(1); + + LogisticRegressionModel model = + new LogisticRegressionModel().setModelData(tEnv.fromDataStream(modelData)); + ReadWriteUtils.updateExistingParams(model, paramMap); + return model; + } + + /** Pre-processes the training data. */ + private static class PreprocessDataOp + extends AbstractUdfStreamOperator< + Tuple3<Double, Double, double[]>, + RichMapFunction< + Tuple3<Double, Double, double[]>, Tuple3<Double, Double, double[]>>> + implements OneInputStreamOperator< + Tuple3<Double, Double, double[]>, Tuple3<Double, Double, double[]>> { + public PreprocessDataOp( + RichMapFunction<Tuple3<Double, Double, double[]>, Tuple3<Double, Double, double[]>> + userFunction) { + super(userFunction); + } + + @Override + public void processElement(StreamRecord<Tuple3<Double, Double, double[]>> streamRecord) + throws Exception { + streamRecord.replace(userFunction.map(streamRecord.getValue())); + output.collect(streamRecord); + } + } + + /** Pre-processes one training sample. */ + private static class PreprocessOneRecord + extends RichMapFunction< + Tuple3<Double, Double, double[]>, Tuple3<Double, Double, double[]>> { + + String broadcastLabelsName; + + double[] distinctLabelValues; + + public PreprocessOneRecord(String broadcastLabelsName) { + this.broadcastLabelsName = broadcastLabelsName; + } + + @Override + public Tuple3<Double, Double, double[]> map(Tuple3<Double, Double, double[]> value) { + if (distinctLabelValues == null) { + List<Double> labelList = + getRuntimeContext().getBroadcastVariable(broadcastLabelsName); + distinctLabelValues = labelList.stream().mapToDouble(Double::doubleValue).toArray(); + } + // label mapping + value.f1 = Math.abs(value.f1 - distinctLabelValues[0]) < 1e-7 ? 1. : -1.; Review comment: Hmm.. instead of hardcoding the label value as either `1` or `-1`, should we choose from the value provided by user? Does Spark's logistic regression takes more than 2 label values? And if we choose to only support two label values, should this value type be boolean? And how is `1e-7` choose? It looks like a magic number which is inconsistent with the `TOLERANCE` used in `LogisticRegressionTest`. Should we make it a `private static final` variable and share it across algorithms? -- 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: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
