lindong28 commented on a change in pull request #54: URL: https://github.com/apache/flink-ml/pull/54#discussion_r829628875
########## File path: flink-ml-lib/src/main/java/org/apache/flink/ml/feature/minmaxscaler/MinMaxScaler.java ########## @@ -0,0 +1,201 @@ +/* + * 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.minmaxscaler; + +import org.apache.flink.api.common.functions.MapFunction; +import org.apache.flink.api.common.functions.RichMapPartitionFunction; +import org.apache.flink.api.common.state.ListState; +import org.apache.flink.api.common.state.ListStateDescriptor; +import org.apache.flink.iteration.operator.OperatorStateUtils; +import org.apache.flink.ml.api.Estimator; +import org.apache.flink.ml.common.datastream.DataStreamUtils; +import org.apache.flink.ml.linalg.DenseVector; +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.environment.StreamExecutionEnvironment; +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.Collector; +import org.apache.flink.util.Preconditions; + +import java.io.IOException; +import java.util.HashMap; +import java.util.Iterator; +import java.util.Map; + +/** + * An Estimator which implements the MinMaxScaler algorithm. + * + * <p>See https://en.wikipedia.org/wiki/Feature_scaling#Rescaling_(min-max_normalization). + */ +public class MinMaxScaler + implements Estimator<MinMaxScaler, MinMaxScalerModel>, MinMaxScalerParams<MinMaxScaler> { + private final Map<Param<?>, Object> paramMap = new HashMap<>(); + + public MinMaxScaler() { + ParamUtils.initializeMapWithDefaultValues(paramMap, this); + } + + @Override + public MinMaxScalerModel fit(Table... inputs) { + Preconditions.checkArgument(inputs.length == 1); + final String featureCol = getFeaturesCol(); + StreamTableEnvironment tEnv = + (StreamTableEnvironment) ((TableImpl) inputs[0]).getTableEnvironment(); + DataStream<DenseVector> features = + tEnv.toDataStream(inputs[0]) + .map( + (MapFunction<Row, DenseVector>) + value -> (DenseVector) value.getField(featureCol)); + DataStream<DenseVector> minMaxValues = + features.transform( + "reduceInEachPartition", + features.getType(), + new MinMaxReduceFunctionOperator()) + .transform( + "reduceInFinalPartition", + features.getType(), + new MinMaxReduceFunctionOperator()) + .setParallelism(1); + DataStream<MinMaxScalerModelData> modelData = + DataStreamUtils.mapPartition( + minMaxValues, + new RichMapPartitionFunction<DenseVector, MinMaxScalerModelData>() { + @Override + public void mapPartition( + Iterable<DenseVector> values, + Collector<MinMaxScalerModelData> out) { + Iterator<DenseVector> iter = values.iterator(); + DenseVector minVector = iter.next(); + DenseVector maxVector = iter.next(); + out.collect(new MinMaxScalerModelData(minVector, maxVector)); + } + }); + + MinMaxScalerModel model = + new MinMaxScalerModel().setModelData(tEnv.fromDataStream(modelData)); + ReadWriteUtils.updateExistingParams(model, getParamMap()); + return model; + } + + /** + * A stream operator to compute the min and max values in each partition of the input bounded + * data stream. + */ + private static class MinMaxReduceFunctionOperator extends AbstractStreamOperator<DenseVector> + implements OneInputStreamOperator<DenseVector, DenseVector>, BoundedOneInput { + private ListState<DenseVector> minState; + private ListState<DenseVector> maxState; + + private DenseVector minVector; + private DenseVector maxVector; + + @Override + public void endInput() { + if (minVector != null) { + output.collect(new StreamRecord<>(minVector)); + } + if (maxVector != null) { + output.collect(new StreamRecord<>(maxVector)); + } + } + + @Override + public void processElement(StreamRecord<DenseVector> streamRecord) { + DenseVector currentValue = streamRecord.getValue(); + if (minVector == null) { + int vecSize = currentValue.size(); + minVector = new DenseVector(vecSize); + maxVector = new DenseVector(vecSize); + System.arraycopy(currentValue.values, 0, minVector.values, 0, vecSize); + System.arraycopy(currentValue.values, 0, maxVector.values, 0, vecSize); + + } else { + for (int i = 0; i < currentValue.size(); ++i) { Review comment: Would it be useful to check that `currentValue.size() = maxVector.size()` here? Maybe follow NaiveBayes.java line 256 for example. ########## File path: flink-ml-lib/src/main/java/org/apache/flink/ml/feature/minmaxscaler/MinMaxScaler.java ########## @@ -0,0 +1,201 @@ +/* + * 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.minmaxscaler; + +import org.apache.flink.api.common.functions.MapFunction; +import org.apache.flink.api.common.functions.RichMapPartitionFunction; +import org.apache.flink.api.common.state.ListState; +import org.apache.flink.api.common.state.ListStateDescriptor; +import org.apache.flink.iteration.operator.OperatorStateUtils; +import org.apache.flink.ml.api.Estimator; +import org.apache.flink.ml.common.datastream.DataStreamUtils; +import org.apache.flink.ml.linalg.DenseVector; +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.environment.StreamExecutionEnvironment; +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.Collector; +import org.apache.flink.util.Preconditions; + +import java.io.IOException; +import java.util.HashMap; +import java.util.Iterator; +import java.util.Map; + +/** + * An Estimator which implements the MinMaxScaler algorithm. + * + * <p>See https://en.wikipedia.org/wiki/Feature_scaling#Rescaling_(min-max_normalization). + */ +public class MinMaxScaler + implements Estimator<MinMaxScaler, MinMaxScalerModel>, MinMaxScalerParams<MinMaxScaler> { + private final Map<Param<?>, Object> paramMap = new HashMap<>(); + + public MinMaxScaler() { + ParamUtils.initializeMapWithDefaultValues(paramMap, this); + } + + @Override + public MinMaxScalerModel fit(Table... inputs) { + Preconditions.checkArgument(inputs.length == 1); + final String featureCol = getFeaturesCol(); + StreamTableEnvironment tEnv = + (StreamTableEnvironment) ((TableImpl) inputs[0]).getTableEnvironment(); + DataStream<DenseVector> features = + tEnv.toDataStream(inputs[0]) + .map( + (MapFunction<Row, DenseVector>) + value -> (DenseVector) value.getField(featureCol)); + DataStream<DenseVector> minMaxValues = + features.transform( + "reduceInEachPartition", + features.getType(), + new MinMaxReduceFunctionOperator()) + .transform( + "reduceInFinalPartition", + features.getType(), + new MinMaxReduceFunctionOperator()) + .setParallelism(1); + DataStream<MinMaxScalerModelData> modelData = + DataStreamUtils.mapPartition( + minMaxValues, + new RichMapPartitionFunction<DenseVector, MinMaxScalerModelData>() { + @Override + public void mapPartition( + Iterable<DenseVector> values, + Collector<MinMaxScalerModelData> out) { + Iterator<DenseVector> iter = values.iterator(); + DenseVector minVector = iter.next(); + DenseVector maxVector = iter.next(); + out.collect(new MinMaxScalerModelData(minVector, maxVector)); + } + }); + + MinMaxScalerModel model = + new MinMaxScalerModel().setModelData(tEnv.fromDataStream(modelData)); + ReadWriteUtils.updateExistingParams(model, getParamMap()); + return model; + } + + /** + * A stream operator to compute the min and max values in each partition of the input bounded + * data stream. + */ + private static class MinMaxReduceFunctionOperator extends AbstractStreamOperator<DenseVector> + implements OneInputStreamOperator<DenseVector, DenseVector>, BoundedOneInput { + private ListState<DenseVector> minState; + private ListState<DenseVector> maxState; + + private DenseVector minVector; + private DenseVector maxVector; + + @Override + public void endInput() { + if (minVector != null) { Review comment: Would it be simpler to just do the following: ``` if (minVector != null) { output.collect(new StreamRecord<>(minVector)); output.collect(new StreamRecord<>(maxVector)); } ``` ########## File path: flink-ml-lib/src/main/java/org/apache/flink/ml/feature/minmaxscaler/MinMaxScaler.java ########## @@ -0,0 +1,201 @@ +/* + * 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.minmaxscaler; + +import org.apache.flink.api.common.functions.MapFunction; +import org.apache.flink.api.common.functions.RichMapPartitionFunction; +import org.apache.flink.api.common.state.ListState; +import org.apache.flink.api.common.state.ListStateDescriptor; +import org.apache.flink.iteration.operator.OperatorStateUtils; +import org.apache.flink.ml.api.Estimator; +import org.apache.flink.ml.common.datastream.DataStreamUtils; +import org.apache.flink.ml.linalg.DenseVector; +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.environment.StreamExecutionEnvironment; +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.Collector; +import org.apache.flink.util.Preconditions; + +import java.io.IOException; +import java.util.HashMap; +import java.util.Iterator; +import java.util.Map; + +/** + * An Estimator which implements the MinMaxScaler algorithm. + * + * <p>See https://en.wikipedia.org/wiki/Feature_scaling#Rescaling_(min-max_normalization). Review comment: Since the link provided here does not explain the use of min/max parameter, it does not exactly describe what this class does. Should we provide more detailed algorithm explanation here similar to Spark's MinMaxScaler Javadoc? ########## File path: flink-ml-lib/src/main/java/org/apache/flink/ml/feature/minmaxscaler/MinMaxScalerModel.java ########## @@ -0,0 +1,181 @@ +/* + * 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.minmaxscaler; + +import org.apache.flink.api.common.functions.RichMapFunction; +import org.apache.flink.api.java.typeutils.RowTypeInfo; +import org.apache.flink.ml.api.Model; +import org.apache.flink.ml.common.broadcast.BroadcastUtils; +import org.apache.flink.ml.common.datastream.TableUtils; +import org.apache.flink.ml.linalg.DenseVector; +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.streaming.api.datastream.DataStream; +import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment; +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.table.runtime.typeutils.ExternalTypeInfo; +import org.apache.flink.types.Row; +import org.apache.flink.util.Preconditions; + +import org.apache.commons.lang3.ArrayUtils; + +import java.io.IOException; +import java.util.Collections; +import java.util.HashMap; +import java.util.Map; + +/** + * A Model which do a minMax scaler operation using the model data computed by {@link MinMaxScaler}. + */ +public class MinMaxScalerModel + implements Model<MinMaxScalerModel>, MinMaxScalerParams<MinMaxScalerModel> { + private final Map<Param<?>, Object> paramMap = new HashMap<>(); + private Table modelDataTable; + + public MinMaxScalerModel() { + ParamUtils.initializeMapWithDefaultValues(paramMap, this); + } + + @Override + public MinMaxScalerModel setModelData(Table... inputs) { + modelDataTable = inputs[0]; + return this; + } + + @Override + public Table[] getModelData() { + return new Table[] {modelDataTable}; + } + + @Override + @SuppressWarnings("unchecked") + public Table[] transform(Table... inputs) { + Preconditions.checkArgument(inputs.length == 1); + StreamTableEnvironment tEnv = + (StreamTableEnvironment) ((TableImpl) inputs[0]).getTableEnvironment(); + DataStream<Row> data = tEnv.toDataStream(inputs[0]); + DataStream<MinMaxScalerModelData> minMaxScalerModel = + MinMaxScalerModelData.getModelDataStream(modelDataTable); + final String broadcastModelKey = "broadcastModelKey"; + RowTypeInfo inputTypeInfo = TableUtils.getRowTypeInfo(inputs[0].getResolvedSchema()); + RowTypeInfo outputTypeInfo = + new RowTypeInfo( + ArrayUtils.addAll( + inputTypeInfo.getFieldTypes(), + ExternalTypeInfo.of(DenseVector.class)), + ArrayUtils.addAll(inputTypeInfo.getFieldNames(), getPredictionCol())); + DataStream<Row> output = + BroadcastUtils.withBroadcastStream( + Collections.singletonList(data), + Collections.singletonMap(broadcastModelKey, minMaxScalerModel), + inputList -> { + DataStream input = inputList.get(0); + return input.map( + new PredictOutputFunction( + broadcastModelKey, + getMax(), + getMin(), + getFeaturesCol()), + outputTypeInfo); + }); + return new Table[] {tEnv.fromDataStream(output)}; + } + + @Override + public Map<Param<?>, Object> getParamMap() { + return paramMap; + } + + @Override + public void save(String path) throws IOException { + ReadWriteUtils.saveMetadata(this, path); + ReadWriteUtils.saveModelData( + MinMaxScalerModelData.getModelDataStream(modelDataTable), + path, + new MinMaxScalerModelData.ModelDataEncoder()); + } + + /** + * Loads model data from path. + * + * @param env Stream execution environment. + * @param path Model path. + * @return MinMaxScalerModel model. + */ + public static MinMaxScalerModel load(StreamExecutionEnvironment env, String path) + throws IOException { + StreamTableEnvironment tEnv = StreamTableEnvironment.create(env); + MinMaxScalerModel model = ReadWriteUtils.loadStageParam(path); + DataStream<MinMaxScalerModelData> modelData = + ReadWriteUtils.loadModelData( + env, path, new MinMaxScalerModelData.ModelDataDecoder()); + return model.setModelData(tEnv.fromDataStream(modelData)); + } + + /** This operator loads model data and predicts result. */ + private static class PredictOutputFunction extends RichMapFunction<Row, Row> { + private final String featureCol; + private MinMaxScalerModelData minMaxScalerModelData; + private final double upperBound; + private final double lowerBound; + private final String broadcastKey; + private DenseVector maxVector; + private DenseVector minVector; + + public PredictOutputFunction( + String broadcastKey, double upperBound, double lowerBound, String featureCol) { + this.upperBound = upperBound; + this.lowerBound = lowerBound; + this.broadcastKey = broadcastKey; + this.featureCol = featureCol; + } + + @Override + public Row map(Row row) { + if (minMaxScalerModelData == null) { + minMaxScalerModelData = + (MinMaxScalerModelData) + getRuntimeContext().getBroadcastVariable(broadcastKey).get(0); + maxVector = minMaxScalerModelData.maxVector; + minVector = minMaxScalerModelData.minVector; + } + DenseVector feature = (DenseVector) row.getField(featureCol); + DenseVector outputVector = new DenseVector(maxVector.size()); Review comment: nits: would it be simpler to rename `outputVector` as `output` for simplicity and consistency with `feature`? -- 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]
