zhipeng93 commented on a change in pull request #54: URL: https://github.com/apache/flink-ml/pull/54#discussion_r823332320
########## File path: flink-ml-lib/src/main/java/org/apache/flink/ml/feature/minmaxscaler/MinMaxScaler.java ########## @@ -0,0 +1,165 @@ +/* + * 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.RichMapPartitionFunction; +import org.apache.flink.api.java.tuple.Tuple2; +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.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.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.Map; + +/** An Estimator which implements the MinMaxScaler algorithm. */ +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); + StreamTableEnvironment tEnv = + (StreamTableEnvironment) ((TableImpl) inputs[0]).getTableEnvironment(); + DataStream<Tuple2<DenseVector, DenseVector>> minMaxVectors = + computeMinMaxVectors(tEnv.toDataStream(inputs[0]), getFeaturesCol()); + DataStream<MinMaxScalerModelData> modelData = genModelData(minMaxVectors); + MinMaxScalerModel model = + new MinMaxScalerModel().setModelData(tEnv.fromDataStream(modelData)); + ReadWriteUtils.updateExistingParams(model, getParamMap()); + return model; + } + + @Override + public Map<Param<?>, Object> getParamMap() { + return paramMap; + } + + @Override + public void save(String path) throws IOException { + ReadWriteUtils.saveMetadata(this, path); + } + + public static MinMaxScaler load(StreamExecutionEnvironment env, String path) + throws IOException { + return ReadWriteUtils.loadStageParam(path); + } + + /** + * Generates minMax scaler model data. + * + * @param minMaxVectors Input distributed minMaxVectors. + * @return MinMax scaler model data. + */ + private static DataStream<MinMaxScalerModelData> genModelData( + DataStream<Tuple2<DenseVector, DenseVector>> minMaxVectors) { + DataStream<MinMaxScalerModelData> modelData = + DataStreamUtils.mapPartition( + minMaxVectors, + new RichMapPartitionFunction< + Tuple2<DenseVector, DenseVector>, MinMaxScalerModelData>() { + @Override + public void mapPartition( + Iterable<Tuple2<DenseVector, DenseVector>> dataPoints, + Collector<MinMaxScalerModelData> out) { + DenseVector minVector = null; + DenseVector maxVector = null; + int vecSize = 0; + for (Tuple2<DenseVector, DenseVector> dataPoint : dataPoints) { + if (maxVector == null) { + vecSize = dataPoint.f0.size(); + maxVector = dataPoint.f1; + minVector = dataPoint.f0; + } + for (int i = 0; i < vecSize; ++i) { Review comment: nits: vecSize could be replaced with `maxVector.size()` ########## File path: flink-ml-lib/src/main/java/org/apache/flink/ml/feature/minmaxscaler/MinMaxScaler.java ########## @@ -0,0 +1,165 @@ +/* + * 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.RichMapPartitionFunction; +import org.apache.flink.api.java.tuple.Tuple2; +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.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.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.Map; + +/** An Estimator which implements the MinMaxScaler algorithm. */ Review comment: Could you add a doc/link to explain what is a `minmaxscaler` as we did for other algorithms? ########## File path: flink-ml-lib/src/main/java/org/apache/flink/ml/feature/minmaxscaler/MinMaxScalerParams.java ########## @@ -0,0 +1,56 @@ +/* + * 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.ml.common.param.HasFeaturesCol; +import org.apache.flink.ml.common.param.HasOutputCol; +import org.apache.flink.ml.param.DoubleParam; +import org.apache.flink.ml.param.Param; +import org.apache.flink.ml.param.ParamValidators; + +/** + * Params for {@link MinMaxScaler}. + * + * @param <T> The class type of this instance. + */ +public interface MinMaxScalerParams<T> extends HasFeaturesCol<T>, HasOutputCol<T> { + Param<Double> MAX = + new DoubleParam( + "max", "Upper bound after transformation.", 1.0, ParamValidators.notNull()); + + default Double getMax() { + return get(MAX); + } + + default T setMax(Double value) { + return set(MAX, value); + } + + Param<Double> MIN = + new DoubleParam( + "min", "Lower bound after transformation.", 0.0, ParamValidators.notNull()); + + default Double getMIN() { Review comment: nit: `getMIN` to `getMin`. ########## File path: flink-ml-lib/src/test/java/org/apache/flink/ml/feature/MinMaxScalerTest.java ########## @@ -0,0 +1,206 @@ +/* + * 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; + +import org.apache.flink.api.common.functions.MapFunction; +import org.apache.flink.api.common.restartstrategy.RestartStrategies; +import org.apache.flink.configuration.Configuration; +import org.apache.flink.ml.feature.minmaxscaler.MinMaxScaler; +import org.apache.flink.ml.feature.minmaxscaler.MinMaxScalerModel; +import org.apache.flink.ml.feature.minmaxscaler.MinMaxScalerModelData; +import org.apache.flink.ml.linalg.DenseVector; +import org.apache.flink.ml.linalg.Vectors; +import org.apache.flink.ml.util.ReadWriteUtils; +import org.apache.flink.ml.util.StageTestUtils; +import org.apache.flink.streaming.api.datastream.DataStream; +import org.apache.flink.streaming.api.environment.ExecutionCheckpointingOptions; +import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment; +import org.apache.flink.table.api.DataTypes; +import org.apache.flink.table.api.Schema; +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.collections.IteratorUtils; +import org.junit.Assert; +import org.junit.Before; +import org.junit.Rule; +import org.junit.Test; +import org.junit.rules.TemporaryFolder; + +import java.util.ArrayList; +import java.util.Arrays; +import java.util.Collections; +import java.util.List; + +import static org.junit.Assert.assertEquals; + +/** Tests {@link MinMaxScaler} and {@link MinMaxScalerModel}. */ +public class MinMaxScalerTest { + @Rule public final TemporaryFolder tempFolder = new TemporaryFolder(); + private StreamExecutionEnvironment env; + private StreamTableEnvironment tEnv; + private Table trainData; + private Table predictData; + private static final List<Row> trainRows = + new ArrayList<>( + Arrays.asList( + Row.of(Vectors.dense(0.0, 3.0)), + Row.of(Vectors.dense(2.1, 0.0)), + Row.of(Vectors.dense(4.1, 5.1)), + Row.of(Vectors.dense(6.1, 8.1)), + Row.of(Vectors.dense(200, 300)))); + private static final List<Row> predictRows = + new ArrayList<>(Collections.singletonList(Row.of(Vectors.dense(150.0, 90.0)))); + + @Before + public void before() { + Configuration config = new Configuration(); + config.set(ExecutionCheckpointingOptions.ENABLE_CHECKPOINTS_AFTER_TASKS_FINISH, true); + env = StreamExecutionEnvironment.getExecutionEnvironment(config); + env.setParallelism(4); + env.enableCheckpointing(100); + env.setRestartStrategy(RestartStrategies.noRestart()); + tEnv = StreamTableEnvironment.create(env); + Schema schema = Schema.newBuilder().column("f0", DataTypes.of(DenseVector.class)).build(); + DataStream<Row> dataStream = env.fromCollection(trainRows); + trainData = tEnv.fromDataStream(dataStream, schema).as("features"); + DataStream<Row> predDataStream = env.fromCollection(predictRows); + predictData = tEnv.fromDataStream(predDataStream, schema).as("features"); + } + + private static void verifyPredictionResult(Table output, String outputCol) throws Exception { + StreamTableEnvironment tEnv = + (StreamTableEnvironment) ((TableImpl) output).getTableEnvironment(); + DataStream<DenseVector> stream = + tEnv.toDataStream(output) + .map( + (MapFunction<Row, DenseVector>) + row -> (DenseVector) row.getField(outputCol)); + List<DenseVector> result = IteratorUtils.toList(stream.executeAndCollect()); + for (DenseVector t2 : result) { + assertEquals(Vectors.dense(0.75, 0.3), t2); + } + } + + @Test + public void testParam() { + MinMaxScaler minMaxScaler = new MinMaxScaler(); + assertEquals("features", minMaxScaler.getFeaturesCol()); + assertEquals(1.0, minMaxScaler.getMax(), 0.0001); + assertEquals(0.0, minMaxScaler.getMIN(), 0.0001); + assertEquals("output", minMaxScaler.getOutputCol()); + minMaxScaler + .setFeaturesCol("test_features") + .setMax(4.0) + .setMIN(0.0) + .setOutputCol("test_output"); + assertEquals("test_features", minMaxScaler.getFeaturesCol()); + assertEquals(0.0, minMaxScaler.getMIN(), 0.0001); + assertEquals(4.0, minMaxScaler.getMax(), 0.0001); + assertEquals("test_output", minMaxScaler.getOutputCol()); + } + + @Test + public void testFeaturePredictionParam() { + MinMaxScaler minMaxScaler = + new MinMaxScaler() + .setMIN(0.0) + .setMax(4.0) + .setFeaturesCol("test_features") + .setOutputCol("test_output"); + MinMaxScalerModel model = minMaxScaler.fit(trainData.as("test_features")); + Table output = model.transform(predictData.as("test_features"))[0]; + assertEquals( + Arrays.asList("test_features", "test_output"), + output.getResolvedSchema().getColumnNames()); + } + + @Test + public void testFewerDistinctPointsThanCluster() { Review comment: Is this test necessary? ########## File path: flink-ml-lib/src/main/java/org/apache/flink/ml/feature/minmaxscaler/MinMaxScaler.java ########## @@ -0,0 +1,165 @@ +/* + * 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.RichMapPartitionFunction; +import org.apache.flink.api.java.tuple.Tuple2; +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.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.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.Map; + +/** An Estimator which implements the MinMaxScaler algorithm. */ +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); + StreamTableEnvironment tEnv = + (StreamTableEnvironment) ((TableImpl) inputs[0]).getTableEnvironment(); + DataStream<Tuple2<DenseVector, DenseVector>> minMaxVectors = + computeMinMaxVectors(tEnv.toDataStream(inputs[0]), getFeaturesCol()); + DataStream<MinMaxScalerModelData> modelData = genModelData(minMaxVectors); + MinMaxScalerModel model = + new MinMaxScalerModel().setModelData(tEnv.fromDataStream(modelData)); + ReadWriteUtils.updateExistingParams(model, getParamMap()); + return model; + } + + @Override + public Map<Param<?>, Object> getParamMap() { + return paramMap; + } + + @Override + public void save(String path) throws IOException { + ReadWriteUtils.saveMetadata(this, path); + } + + public static MinMaxScaler load(StreamExecutionEnvironment env, String path) + throws IOException { + return ReadWriteUtils.loadStageParam(path); + } + + /** + * Generates minMax scaler model data. + * + * @param minMaxVectors Input distributed minMaxVectors. + * @return MinMax scaler model data. + */ + private static DataStream<MinMaxScalerModelData> genModelData( + DataStream<Tuple2<DenseVector, DenseVector>> minMaxVectors) { + DataStream<MinMaxScalerModelData> modelData = + DataStreamUtils.mapPartition( + minMaxVectors, + new RichMapPartitionFunction< + Tuple2<DenseVector, DenseVector>, MinMaxScalerModelData>() { + @Override + public void mapPartition( + Iterable<Tuple2<DenseVector, DenseVector>> dataPoints, + Collector<MinMaxScalerModelData> out) { + DenseVector minVector = null; + DenseVector maxVector = null; + int vecSize = 0; + for (Tuple2<DenseVector, DenseVector> dataPoint : dataPoints) { + if (maxVector == null) { + vecSize = dataPoint.f0.size(); + maxVector = dataPoint.f1; + minVector = dataPoint.f0; + } + for (int i = 0; i < vecSize; ++i) { + minVector.values[i] = + Math.min( + dataPoint.f0.values[i], + minVector.values[i]); + maxVector.values[i] = + Math.max( + dataPoint.f1.values[i], + maxVector.values[i]); + } + } + out.collect(new MinMaxScalerModelData(minVector, maxVector)); + } + }); + modelData.getTransformation().setParallelism(1); + return modelData; + } + + /** + * Computes max and min values of features. + * + * @param inputData Input data. + * @param featureCol Feature column name. + * @return Max and min values of features. + */ + private DataStream<Tuple2<DenseVector, DenseVector>> computeMinMaxVectors( Review comment: The logic of `computeMinMaxVectors` and `genModelData` seems almost the same. Is merging them into one a better implementation? ########## File path: flink-ml-lib/src/main/java/org/apache/flink/ml/feature/minmaxscaler/MinMaxScalerModel.java ########## @@ -0,0 +1,179 @@ +/* + * 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(), getOutputCol())); + DataStream<Row> output = + BroadcastUtils.withBroadcastStream( + Collections.singletonList(data), + Collections.singletonMap(broadcastModelKey, minMaxScalerModel), + inputList -> { + DataStream input = inputList.get(0); + return input.map( + new PredictLabelFunction( + 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 PredictLabelFunction extends RichMapFunction<Row, Row> { + private final String featureCol; + private MinMaxScalerModelData minMaxScalerModelData; + private final double max; + private final double min; + private final String broadcastKey; + private DenseVector maxVector; + private DenseVector minVector; + + public PredictLabelFunction( + String broadcastKey, double max, double min, String featureCol) { + this.max = max; Review comment: nit: Are `loweBound` and `upperBound` better names for `max` and `min`? ########## File path: flink-ml-lib/src/main/java/org/apache/flink/ml/common/param/HasOutputCol.java ########## @@ -0,0 +1,39 @@ +/* + * 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.param; + +import org.apache.flink.ml.param.Param; +import org.apache.flink.ml.param.ParamValidators; +import org.apache.flink.ml.param.StringParam; +import org.apache.flink.ml.param.WithParams; + +/** Interface for the shared outputCol param. */ +public interface HasOutputCol<T> extends WithParams<T> { Review comment: Why do we need a new param here? How about using the existing one, i.e., `HasPredictionCol`? ########## File path: flink-ml-lib/src/main/java/org/apache/flink/ml/feature/minmaxscaler/MinMaxScaler.java ########## @@ -0,0 +1,165 @@ +/* + * 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.RichMapPartitionFunction; +import org.apache.flink.api.java.tuple.Tuple2; +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.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.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.Map; + +/** An Estimator which implements the MinMaxScaler algorithm. */ +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); + StreamTableEnvironment tEnv = + (StreamTableEnvironment) ((TableImpl) inputs[0]).getTableEnvironment(); + DataStream<Tuple2<DenseVector, DenseVector>> minMaxVectors = + computeMinMaxVectors(tEnv.toDataStream(inputs[0]), getFeaturesCol()); + DataStream<MinMaxScalerModelData> modelData = genModelData(minMaxVectors); + MinMaxScalerModel model = + new MinMaxScalerModel().setModelData(tEnv.fromDataStream(modelData)); + ReadWriteUtils.updateExistingParams(model, getParamMap()); + return model; + } + + @Override + public Map<Param<?>, Object> getParamMap() { + return paramMap; + } + + @Override + public void save(String path) throws IOException { + ReadWriteUtils.saveMetadata(this, path); + } + + public static MinMaxScaler load(StreamExecutionEnvironment env, String path) + throws IOException { + return ReadWriteUtils.loadStageParam(path); + } + + /** + * Generates minMax scaler model data. + * + * @param minMaxVectors Input distributed minMaxVectors. + * @return MinMax scaler model data. + */ + private static DataStream<MinMaxScalerModelData> genModelData( + DataStream<Tuple2<DenseVector, DenseVector>> minMaxVectors) { + DataStream<MinMaxScalerModelData> modelData = + DataStreamUtils.mapPartition( Review comment: nits: using `mapPartition` here may be not that efficient since we cache all the input records in the state. How about using a self-defined streamOperator as [1]? [1] https://github.com/zhipeng93/flink-ml/blob/ccd3f9919dbdd3c5123f1cc88636f832f8215ffb/flink-ml-core/src/main/java/org/apache/flink/ml/common/datastream/DataStreamUtils.java#L68 ########## File path: flink-ml-lib/src/test/java/org/apache/flink/ml/feature/MinMaxScalerTest.java ########## @@ -0,0 +1,206 @@ +/* + * 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; + +import org.apache.flink.api.common.functions.MapFunction; +import org.apache.flink.api.common.restartstrategy.RestartStrategies; +import org.apache.flink.configuration.Configuration; +import org.apache.flink.ml.feature.minmaxscaler.MinMaxScaler; +import org.apache.flink.ml.feature.minmaxscaler.MinMaxScalerModel; +import org.apache.flink.ml.feature.minmaxscaler.MinMaxScalerModelData; +import org.apache.flink.ml.linalg.DenseVector; +import org.apache.flink.ml.linalg.Vectors; +import org.apache.flink.ml.util.ReadWriteUtils; +import org.apache.flink.ml.util.StageTestUtils; +import org.apache.flink.streaming.api.datastream.DataStream; +import org.apache.flink.streaming.api.environment.ExecutionCheckpointingOptions; +import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment; +import org.apache.flink.table.api.DataTypes; +import org.apache.flink.table.api.Schema; +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.collections.IteratorUtils; +import org.junit.Assert; +import org.junit.Before; +import org.junit.Rule; +import org.junit.Test; +import org.junit.rules.TemporaryFolder; + +import java.util.ArrayList; +import java.util.Arrays; +import java.util.Collections; +import java.util.List; + +import static org.junit.Assert.assertEquals; + +/** Tests {@link MinMaxScaler} and {@link MinMaxScalerModel}. */ +public class MinMaxScalerTest { + @Rule public final TemporaryFolder tempFolder = new TemporaryFolder(); + private StreamExecutionEnvironment env; + private StreamTableEnvironment tEnv; + private Table trainData; + private Table predictData; + private static final List<Row> trainRows = + new ArrayList<>( + Arrays.asList( + Row.of(Vectors.dense(0.0, 3.0)), + Row.of(Vectors.dense(2.1, 0.0)), + Row.of(Vectors.dense(4.1, 5.1)), + Row.of(Vectors.dense(6.1, 8.1)), + Row.of(Vectors.dense(200, 300)))); + private static final List<Row> predictRows = + new ArrayList<>(Collections.singletonList(Row.of(Vectors.dense(150.0, 90.0)))); + + @Before + public void before() { + Configuration config = new Configuration(); + config.set(ExecutionCheckpointingOptions.ENABLE_CHECKPOINTS_AFTER_TASKS_FINISH, true); + env = StreamExecutionEnvironment.getExecutionEnvironment(config); + env.setParallelism(4); + env.enableCheckpointing(100); + env.setRestartStrategy(RestartStrategies.noRestart()); + tEnv = StreamTableEnvironment.create(env); + Schema schema = Schema.newBuilder().column("f0", DataTypes.of(DenseVector.class)).build(); + DataStream<Row> dataStream = env.fromCollection(trainRows); + trainData = tEnv.fromDataStream(dataStream, schema).as("features"); + DataStream<Row> predDataStream = env.fromCollection(predictRows); + predictData = tEnv.fromDataStream(predDataStream, schema).as("features"); + } + + private static void verifyPredictionResult(Table output, String outputCol) throws Exception { + StreamTableEnvironment tEnv = + (StreamTableEnvironment) ((TableImpl) output).getTableEnvironment(); + DataStream<DenseVector> stream = + tEnv.toDataStream(output) + .map( + (MapFunction<Row, DenseVector>) + row -> (DenseVector) row.getField(outputCol)); + List<DenseVector> result = IteratorUtils.toList(stream.executeAndCollect()); + for (DenseVector t2 : result) { Review comment: Is this loop necessary here? ########## File path: flink-ml-lib/src/test/java/org/apache/flink/ml/feature/MinMaxScalerTest.java ########## @@ -0,0 +1,206 @@ +/* + * 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; + +import org.apache.flink.api.common.functions.MapFunction; +import org.apache.flink.api.common.restartstrategy.RestartStrategies; +import org.apache.flink.configuration.Configuration; +import org.apache.flink.ml.feature.minmaxscaler.MinMaxScaler; +import org.apache.flink.ml.feature.minmaxscaler.MinMaxScalerModel; +import org.apache.flink.ml.feature.minmaxscaler.MinMaxScalerModelData; +import org.apache.flink.ml.linalg.DenseVector; +import org.apache.flink.ml.linalg.Vectors; +import org.apache.flink.ml.util.ReadWriteUtils; +import org.apache.flink.ml.util.StageTestUtils; +import org.apache.flink.streaming.api.datastream.DataStream; +import org.apache.flink.streaming.api.environment.ExecutionCheckpointingOptions; +import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment; +import org.apache.flink.table.api.DataTypes; +import org.apache.flink.table.api.Schema; +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.collections.IteratorUtils; +import org.junit.Assert; +import org.junit.Before; +import org.junit.Rule; +import org.junit.Test; +import org.junit.rules.TemporaryFolder; + +import java.util.ArrayList; +import java.util.Arrays; +import java.util.Collections; +import java.util.List; + +import static org.junit.Assert.assertEquals; + +/** Tests {@link MinMaxScaler} and {@link MinMaxScalerModel}. */ +public class MinMaxScalerTest { + @Rule public final TemporaryFolder tempFolder = new TemporaryFolder(); + private StreamExecutionEnvironment env; + private StreamTableEnvironment tEnv; + private Table trainData; + private Table predictData; + private static final List<Row> trainRows = + new ArrayList<>( + Arrays.asList( + Row.of(Vectors.dense(0.0, 3.0)), + Row.of(Vectors.dense(2.1, 0.0)), + Row.of(Vectors.dense(4.1, 5.1)), + Row.of(Vectors.dense(6.1, 8.1)), + Row.of(Vectors.dense(200, 300)))); + private static final List<Row> predictRows = + new ArrayList<>(Collections.singletonList(Row.of(Vectors.dense(150.0, 90.0)))); + + @Before + public void before() { + Configuration config = new Configuration(); + config.set(ExecutionCheckpointingOptions.ENABLE_CHECKPOINTS_AFTER_TASKS_FINISH, true); + env = StreamExecutionEnvironment.getExecutionEnvironment(config); + env.setParallelism(4); + env.enableCheckpointing(100); + env.setRestartStrategy(RestartStrategies.noRestart()); + tEnv = StreamTableEnvironment.create(env); + Schema schema = Schema.newBuilder().column("f0", DataTypes.of(DenseVector.class)).build(); + DataStream<Row> dataStream = env.fromCollection(trainRows); + trainData = tEnv.fromDataStream(dataStream, schema).as("features"); + DataStream<Row> predDataStream = env.fromCollection(predictRows); + predictData = tEnv.fromDataStream(predDataStream, schema).as("features"); + } + + private static void verifyPredictionResult(Table output, String outputCol) throws Exception { + StreamTableEnvironment tEnv = + (StreamTableEnvironment) ((TableImpl) output).getTableEnvironment(); + DataStream<DenseVector> stream = + tEnv.toDataStream(output) + .map( + (MapFunction<Row, DenseVector>) + row -> (DenseVector) row.getField(outputCol)); + List<DenseVector> result = IteratorUtils.toList(stream.executeAndCollect()); + for (DenseVector t2 : result) { + assertEquals(Vectors.dense(0.75, 0.3), t2); + } + } + + @Test + public void testParam() { + MinMaxScaler minMaxScaler = new MinMaxScaler(); + assertEquals("features", minMaxScaler.getFeaturesCol()); + assertEquals(1.0, minMaxScaler.getMax(), 0.0001); + assertEquals(0.0, minMaxScaler.getMIN(), 0.0001); + assertEquals("output", minMaxScaler.getOutputCol()); + minMaxScaler + .setFeaturesCol("test_features") + .setMax(4.0) + .setMIN(0.0) Review comment: How about use different default value for `min`? ########## File path: flink-ml-lib/src/main/java/org/apache/flink/ml/feature/minmaxscaler/MinMaxScalerParams.java ########## @@ -0,0 +1,56 @@ +/* + * 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.ml.common.param.HasFeaturesCol; +import org.apache.flink.ml.common.param.HasOutputCol; +import org.apache.flink.ml.param.DoubleParam; +import org.apache.flink.ml.param.Param; +import org.apache.flink.ml.param.ParamValidators; + +/** + * Params for {@link MinMaxScaler}. + * + * @param <T> The class type of this instance. + */ +public interface MinMaxScalerParams<T> extends HasFeaturesCol<T>, HasOutputCol<T> { + Param<Double> MAX = + new DoubleParam( + "max", "Upper bound after transformation.", 1.0, ParamValidators.notNull()); + + default Double getMax() { + return get(MAX); + } + + default T setMax(Double value) { + return set(MAX, value); + } + + Param<Double> MIN = + new DoubleParam( + "min", "Lower bound after transformation.", 0.0, ParamValidators.notNull()); + + default Double getMIN() { + return get(MIN); + } + + default T setMIN(Double value) { Review comment: ditto. ########## File path: flink-ml-lib/src/main/java/org/apache/flink/ml/feature/minmaxscaler/MinMaxScalerModel.java ########## @@ -0,0 +1,179 @@ +/* + * 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(), getOutputCol())); + DataStream<Row> output = + BroadcastUtils.withBroadcastStream( + Collections.singletonList(data), + Collections.singletonMap(broadcastModelKey, minMaxScalerModel), + inputList -> { + DataStream input = inputList.get(0); + return input.map( + new PredictLabelFunction( + 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 PredictLabelFunction extends RichMapFunction<Row, Row> { + private final String featureCol; + private MinMaxScalerModelData minMaxScalerModelData; + private final double max; + private final double min; + private final String broadcastKey; + private DenseVector maxVector; + private DenseVector minVector; + + public PredictLabelFunction( + String broadcastKey, double max, double min, String featureCol) { + this.max = max; + this.min = min; + 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()); + if (feature != null) { Review comment: Should we throw an exception here if the input data does not contain the `featureCol`? -- 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
