yunfengzhou-hub commented on a change in pull request #70: URL: https://github.com/apache/flink-ml/pull/70#discussion_r829714815
########## File path: flink-ml-lib/src/main/java/org/apache/flink/ml/clustering/kmeans/StreamingKMeans.java ########## @@ -0,0 +1,404 @@ +/* + * 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.clustering.kmeans; + +import org.apache.flink.api.common.functions.AggregateFunction; +import org.apache.flink.api.common.functions.MapFunction; +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.typeutils.ObjectArrayTypeInfo; +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.Iterations; +import org.apache.flink.ml.api.Estimator; +import org.apache.flink.ml.common.distance.DistanceMeasure; +import org.apache.flink.ml.common.param.HasBatchStrategy; +import org.apache.flink.ml.linalg.BLAS; +import org.apache.flink.ml.linalg.DenseVector; +import org.apache.flink.ml.linalg.Vectors; +import org.apache.flink.ml.linalg.typeinfo.DenseVectorTypeInfo; +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.streaming.api.datastream.DataStream; +import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment; +import org.apache.flink.streaming.api.operators.AbstractStreamOperator; +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.bridge.java.internal.StreamTableEnvironmentImpl; +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 org.apache.commons.lang3.ArrayUtils; + +import java.io.IOException; +import java.nio.file.Files; +import java.nio.file.Path; +import java.nio.file.Paths; +import java.util.ArrayList; +import java.util.HashMap; +import java.util.List; +import java.util.Map; +import java.util.Random; + +/** + * StreamingKMeans extends the function of {@link KMeans}, supporting to train a K-Means model + * continuously according to an unbounded stream of train data. + */ +public class StreamingKMeans + implements Estimator<StreamingKMeans, StreamingKMeansModel>, + StreamingKMeansParams<StreamingKMeans> { + private final Map<Param<?>, Object> paramMap = new HashMap<>(); + private Table initModelDataTable; + + public StreamingKMeans() { + ParamUtils.initializeMapWithDefaultValues(paramMap, this); + } + + public StreamingKMeans(Table... initModelDataTables) { + Preconditions.checkArgument(initModelDataTables.length == 1); + this.initModelDataTable = initModelDataTables[0]; + ParamUtils.initializeMapWithDefaultValues(paramMap, this); + setInitMode("direct"); + } + + @Override + public StreamingKMeansModel fit(Table... inputs) { + Preconditions.checkArgument(inputs.length == 1); + Preconditions.checkArgument(HasBatchStrategy.COUNT_STRATEGY.equals(getBatchStrategy())); + + StreamTableEnvironment tEnv = + (StreamTableEnvironment) ((TableImpl) inputs[0]).getTableEnvironment(); + StreamExecutionEnvironment env = ((StreamTableEnvironmentImpl) tEnv).execEnv(); + + DataStream<DenseVector> points = + tEnv.toDataStream(inputs[0]).map(new FeaturesExtractor(getFeaturesCol())); + points.getTransformation().setParallelism(1); Review comment: I think it is hard to achieve for now. We need to create mini batches of fixed batch size from train data, but if the parallelism is larger than 1, we do not have a mechanism to count the total number of records received by each subtask. One possible solution I have wanted to propose is to insert barrier into the train data stream, so that even if train data would be distributed on different subtasks, the subtasks still knows when to finish the current batch so long as it can receive barrier. We have not got a change to discuss this problem and possible solutions offline. For now I prefer to still have this limit in this PR. I'll add relevant notices in its Javadoc. -- 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]
