Fanoid commented on code in PR #210: URL: https://github.com/apache/flink-ml/pull/210#discussion_r1113988672
########## flink-ml-lib/src/main/java/org/apache/flink/ml/common/gbt/GBTRunner.java: ########## @@ -0,0 +1,247 @@ +/* + * 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.common.gbt; + +import org.apache.flink.api.common.functions.AggregateFunction; +import org.apache.flink.api.common.functions.RichMapFunction; +import org.apache.flink.api.java.tuple.Tuple2; +import org.apache.flink.iteration.DataStreamList; +import org.apache.flink.iteration.IterationConfig; +import org.apache.flink.iteration.IterationID; +import org.apache.flink.iteration.Iterations; +import org.apache.flink.iteration.ReplayableDataStreamList; +import org.apache.flink.ml.classification.gbtclassifier.GBTClassifierParams; +import org.apache.flink.ml.common.broadcast.BroadcastUtils; +import org.apache.flink.ml.common.datastream.DataStreamUtils; +import org.apache.flink.ml.common.gbt.defs.FeatureMeta; +import org.apache.flink.ml.common.gbt.defs.GbtParams; +import org.apache.flink.ml.common.gbt.defs.TaskType; +import org.apache.flink.ml.common.gbt.defs.TrainContext; +import org.apache.flink.ml.param.Param; +import org.apache.flink.ml.regression.gbtregressor.GBTRegressorParams; +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.commons.lang3.ArrayUtils; + +import java.util.Arrays; +import java.util.Collections; +import java.util.Comparator; +import java.util.HashMap; +import java.util.HashSet; +import java.util.List; +import java.util.Map; +import java.util.Set; +import java.util.stream.Collectors; + +/** Runs a gradient boosting trees implementation. */ +public class GBTRunner { + + public static DataStream<GBTModelData> trainClassifier(Table data, BaseGBTParams<?> estimator) { + return train(data, estimator, TaskType.CLASSIFICATION); + } + + public static DataStream<GBTModelData> trainRegressor(Table data, BaseGBTParams<?> estimator) { + return train(data, estimator, TaskType.REGRESSION); + } + + static DataStream<GBTModelData> train( + Table data, BaseGBTParams<?> estimator, TaskType taskType) { + return train(data, fromEstimator(estimator, taskType)); + } + + /** Trains a gradient boosting tree model with given data and parameters. */ + static DataStream<GBTModelData> train(Table dataTable, GbtParams p) { + StreamTableEnvironment tEnv = + (StreamTableEnvironment) ((TableImpl) dataTable).getTableEnvironment(); + Tuple2<Table, DataStream<FeatureMeta>> preprocessResult = + p.isInputVector + ? Preprocess.preprocessVecCol(dataTable, p) + : Preprocess.preprocessCols(dataTable, p); + dataTable = preprocessResult.f0; + DataStream<FeatureMeta> featureMeta = preprocessResult.f1; + + DataStream<Row> data = tEnv.toDataStream(dataTable); + DataStream<Tuple2<Double, Long>> labelSumCount = + DataStreamUtils.aggregate(data, new LabelSumCountFunction(p.labelCol)); + return boost(dataTable, p, featureMeta, labelSumCount); + } + + private static DataStream<GBTModelData> boost( + Table dataTable, + GbtParams p, + DataStream<FeatureMeta> featureMeta, + DataStream<Tuple2<Double, Long>> labelSumCount) { + StreamTableEnvironment tEnv = + (StreamTableEnvironment) ((TableImpl) dataTable).getTableEnvironment(); + + final String featureMetaBcName = "featureMeta"; + final String labelSumCountBcName = "labelSumCount"; + Map<String, DataStream<?>> bcMap = new HashMap<>(); + bcMap.put(featureMetaBcName, featureMeta); + bcMap.put(labelSumCountBcName, labelSumCount); + + DataStream<TrainContext> initTrainContext = + BroadcastUtils.withBroadcastStream( + Collections.singletonList( + tEnv.toDataStream(tEnv.fromValues(0), Integer.class)), + bcMap, + (inputs) -> { + //noinspection unchecked + DataStream<Integer> input = (DataStream<Integer>) (inputs.get(0)); + return input.map( + new InitTrainContextFunction( + featureMetaBcName, labelSumCountBcName, p)); + }); + + DataStream<Row> data = tEnv.toDataStream(dataTable); + final IterationID iterationID = new IterationID(); + DataStreamList dataStreamList = + Iterations.iterateBoundedStreamsUntilTermination( + DataStreamList.of(initTrainContext.broadcast()), + ReplayableDataStreamList.notReplay(data, featureMeta), + IterationConfig.newBuilder() + .setOperatorLifeCycle(IterationConfig.OperatorLifeCycle.ALL_ROUND) + .build(), + new BoostIterationBody(iterationID, p)); + return dataStreamList.get(0); + } + + public static GbtParams fromEstimator(BaseGBTParams<?> estimator, TaskType taskType) { Review Comment: SparkML's implementation has a class named `BoostingStrategy` [1] serving as a container like `GbtParams`. I think its mainly purpose is to make training functions in `GBTRunner` able to be tested separately. [1] https://github.com/apache/spark/blob/27ed89b7be5ebb91e4a0b106b1669a7867a6012d/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala#L195 -- 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
