You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by GitBox <gi...@apache.org> on 2022/07/21 01:42:54 UTC
[GitHub] [flink-ml] yunfengzhou-hub commented on a diff in pull request #133: [FLINK-28601] Support FeatureHasher in FlinkML
yunfengzhou-hub commented on code in PR #133:
URL: https://github.com/apache/flink-ml/pull/133#discussion_r926165477
##########
flink-ml-lib/src/main/java/org/apache/flink/ml/feature/featurehasher/FeatureHasherParams.java:
##########
@@ -0,0 +1,51 @@
+/*
+ * 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.featurehasher;
+
+import org.apache.flink.ml.common.param.HasCategoricalCols;
+import org.apache.flink.ml.common.param.HasInputCols;
+import org.apache.flink.ml.common.param.HasOutputCol;
+import org.apache.flink.ml.param.IntParam;
+import org.apache.flink.ml.param.Param;
+import org.apache.flink.ml.param.ParamValidators;
+
+/**
+ * Params of FeatureHasher.
+ *
+ * @param <T> The class type of this instance.
+ */
+public interface FeatureHasherParams<T>
+ extends HasInputCols<T>, HasOutputCol<T>, HasCategoricalCols<T> {
+
+ Param<Integer> NUM_FEATURES =
+ new IntParam(
+ "numFeatures",
+ "The number of features. It will be the length of the output vector.",
+ 1 << 18,
Review Comment:
The default value for `numFeatures` is `262144`. Is this value or `1<<18` has any special meaning?
##########
flink-ml-lib/src/main/java/org/apache/flink/ml/feature/featurehasher/FeatureHasher.java:
##########
@@ -0,0 +1,143 @@
+/*
+ * 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.featurehasher;
+
+import org.apache.flink.api.common.functions.MapFunction;
+import org.apache.flink.api.java.typeutils.RowTypeInfo;
+import org.apache.flink.ml.api.Transformer;
+import org.apache.flink.ml.common.datastream.TableUtils;
+import org.apache.flink.ml.linalg.SparseVector;
+import org.apache.flink.ml.linalg.typeinfo.VectorTypeInfo;
+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.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.Preconditions;
+
+import org.apache.flink.shaded.guava30.com.google.common.hash.HashFunction;
+
+import org.apache.commons.lang3.ArrayUtils;
+
+import java.io.IOException;
+import java.util.HashMap;
+import java.util.Map;
+import java.util.TreeMap;
+
+import static org.apache.flink.shaded.guava30.com.google.common.hash.Hashing.murmur3_32;
+
+/**
+ * FeatureHasher projects a set of categorical or numerical features into a vector with fixed
+ * dimension. This is done using the hashing trick (https://en.wikipedia.org/wiki/Feature_hashing)
+ * to map features to indices in the feature vector.
+ */
+public class FeatureHasher
+ implements Transformer<FeatureHasher>, FeatureHasherParams<FeatureHasher> {
+ private final Map<Param<?>, Object> paramMap = new HashMap<>();
+ private static final HashFunction HASH = murmur3_32(0);
+
+ public FeatureHasher() {
+ ParamUtils.initializeMapWithDefaultValues(paramMap, this);
+ }
+
+ @Override
+ public Table[] transform(Table... inputs) {
+ Preconditions.checkArgument(inputs.length == 1);
+ StreamTableEnvironment tEnv =
+ (StreamTableEnvironment) ((TableImpl) inputs[0]).getTableEnvironment();
+ RowTypeInfo inputTypeInfo = TableUtils.getRowTypeInfo(inputs[0].getResolvedSchema());
+ RowTypeInfo outputTypeInfo =
+ new RowTypeInfo(
+ ArrayUtils.addAll(inputTypeInfo.getFieldTypes(), VectorTypeInfo.INSTANCE),
+ ArrayUtils.addAll(inputTypeInfo.getFieldNames(), getOutputCol()));
+ DataStream<Row> output =
+ tEnv.toDataStream(inputs[0])
+ .map(
+ new HashFunc(
+ getInputCols(), getCategoricalCols(), getNumFeatures()),
+ outputTypeInfo);
+ Table outputTable = tEnv.fromDataStream(output);
+ return new Table[] {outputTable};
+ }
+
+ private static class HashFunc implements MapFunction<Row, Row> {
+ private final String[] categoricalCols;
+ private final int numFeatures;
+ private final String[] numericCols;
+
+ public HashFunc(String[] inputCols, String[] handleInvalid, int numFeatures) {
+ this.categoricalCols = handleInvalid;
Review Comment:
It might be better to rename `handleInvalid` to `categoricalCols`.
##########
flink-ml-lib/src/main/java/org/apache/flink/ml/feature/featurehasher/FeatureHasher.java:
##########
@@ -0,0 +1,143 @@
+/*
+ * 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.featurehasher;
+
+import org.apache.flink.api.common.functions.MapFunction;
+import org.apache.flink.api.java.typeutils.RowTypeInfo;
+import org.apache.flink.ml.api.Transformer;
+import org.apache.flink.ml.common.datastream.TableUtils;
+import org.apache.flink.ml.linalg.SparseVector;
+import org.apache.flink.ml.linalg.typeinfo.VectorTypeInfo;
+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.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.Preconditions;
+
+import org.apache.flink.shaded.guava30.com.google.common.hash.HashFunction;
+
+import org.apache.commons.lang3.ArrayUtils;
+
+import java.io.IOException;
+import java.util.HashMap;
+import java.util.Map;
+import java.util.TreeMap;
+
+import static org.apache.flink.shaded.guava30.com.google.common.hash.Hashing.murmur3_32;
+
+/**
+ * FeatureHasher projects a set of categorical or numerical features into a vector with fixed
+ * dimension. This is done using the hashing trick (https://en.wikipedia.org/wiki/Feature_hashing)
+ * to map features to indices in the feature vector.
+ */
+public class FeatureHasher
+ implements Transformer<FeatureHasher>, FeatureHasherParams<FeatureHasher> {
+ private final Map<Param<?>, Object> paramMap = new HashMap<>();
+ private static final HashFunction HASH = murmur3_32(0);
+
+ public FeatureHasher() {
+ ParamUtils.initializeMapWithDefaultValues(paramMap, this);
+ }
+
+ @Override
+ public Table[] transform(Table... inputs) {
+ Preconditions.checkArgument(inputs.length == 1);
Review Comment:
It might be better to check it whether `categoricalCols` are a subarray of `inputCols`.
##########
flink-ml-lib/src/main/java/org/apache/flink/ml/feature/featurehasher/FeatureHasher.java:
##########
@@ -0,0 +1,143 @@
+/*
+ * 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.featurehasher;
+
+import org.apache.flink.api.common.functions.MapFunction;
+import org.apache.flink.api.java.typeutils.RowTypeInfo;
+import org.apache.flink.ml.api.Transformer;
+import org.apache.flink.ml.common.datastream.TableUtils;
+import org.apache.flink.ml.linalg.SparseVector;
+import org.apache.flink.ml.linalg.typeinfo.VectorTypeInfo;
+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.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.Preconditions;
+
+import org.apache.flink.shaded.guava30.com.google.common.hash.HashFunction;
+
+import org.apache.commons.lang3.ArrayUtils;
+
+import java.io.IOException;
+import java.util.HashMap;
+import java.util.Map;
+import java.util.TreeMap;
+
+import static org.apache.flink.shaded.guava30.com.google.common.hash.Hashing.murmur3_32;
+
+/**
+ * FeatureHasher projects a set of categorical or numerical features into a vector with fixed
+ * dimension. This is done using the hashing trick (https://en.wikipedia.org/wiki/Feature_hashing)
+ * to map features to indices in the feature vector.
+ */
+public class FeatureHasher
+ implements Transformer<FeatureHasher>, FeatureHasherParams<FeatureHasher> {
+ private final Map<Param<?>, Object> paramMap = new HashMap<>();
+ private static final HashFunction HASH = murmur3_32(0);
+
+ public FeatureHasher() {
+ ParamUtils.initializeMapWithDefaultValues(paramMap, this);
+ }
+
+ @Override
+ public Table[] transform(Table... inputs) {
+ Preconditions.checkArgument(inputs.length == 1);
+ StreamTableEnvironment tEnv =
+ (StreamTableEnvironment) ((TableImpl) inputs[0]).getTableEnvironment();
+ RowTypeInfo inputTypeInfo = TableUtils.getRowTypeInfo(inputs[0].getResolvedSchema());
+ RowTypeInfo outputTypeInfo =
+ new RowTypeInfo(
+ ArrayUtils.addAll(inputTypeInfo.getFieldTypes(), VectorTypeInfo.INSTANCE),
+ ArrayUtils.addAll(inputTypeInfo.getFieldNames(), getOutputCol()));
+ DataStream<Row> output =
+ tEnv.toDataStream(inputs[0])
+ .map(
+ new HashFunc(
+ getInputCols(), getCategoricalCols(), getNumFeatures()),
Review Comment:
Users should be allowed not to set `categoricalCols`, if they want all input columns to b numerical. For now, if users don't set this parameter, the `getCategoricalCols()` here would throw an exception.
##########
flink-ml-lib/src/main/java/org/apache/flink/ml/feature/featurehasher/FeatureHasherParams.java:
##########
@@ -0,0 +1,51 @@
+/*
+ * 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.featurehasher;
+
+import org.apache.flink.ml.common.param.HasCategoricalCols;
+import org.apache.flink.ml.common.param.HasInputCols;
+import org.apache.flink.ml.common.param.HasOutputCol;
+import org.apache.flink.ml.param.IntParam;
+import org.apache.flink.ml.param.Param;
+import org.apache.flink.ml.param.ParamValidators;
+
+/**
+ * Params of FeatureHasher.
+ *
+ * @param <T> The class type of this instance.
+ */
+public interface FeatureHasherParams<T>
+ extends HasInputCols<T>, HasOutputCol<T>, HasCategoricalCols<T> {
Review Comment:
Would it be better to make `categoricalCol` an exclusive parameter of FeatureHasher, instead of making it a common parameter? Would this parameter be used in other algorithms as well?
Besides, `numFeatures` is shared between `FeatureHasher` and HashingTF`, so maybe we should make this parameter a common parameter. I'm also OK with it if we keep it as an exclusive parameter for now.
##########
flink-ml-lib/src/main/java/org/apache/flink/ml/feature/featurehasher/FeatureHasher.java:
##########
@@ -0,0 +1,143 @@
+/*
+ * 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.featurehasher;
+
+import org.apache.flink.api.common.functions.MapFunction;
+import org.apache.flink.api.java.typeutils.RowTypeInfo;
+import org.apache.flink.ml.api.Transformer;
+import org.apache.flink.ml.common.datastream.TableUtils;
+import org.apache.flink.ml.linalg.SparseVector;
+import org.apache.flink.ml.linalg.typeinfo.VectorTypeInfo;
+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.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.Preconditions;
+
+import org.apache.flink.shaded.guava30.com.google.common.hash.HashFunction;
+
+import org.apache.commons.lang3.ArrayUtils;
+
+import java.io.IOException;
+import java.util.HashMap;
+import java.util.Map;
+import java.util.TreeMap;
+
+import static org.apache.flink.shaded.guava30.com.google.common.hash.Hashing.murmur3_32;
+
+/**
+ * FeatureHasher projects a set of categorical or numerical features into a vector with fixed
+ * dimension. This is done using the hashing trick (https://en.wikipedia.org/wiki/Feature_hashing)
+ * to map features to indices in the feature vector.
+ */
+public class FeatureHasher
+ implements Transformer<FeatureHasher>, FeatureHasherParams<FeatureHasher> {
+ private final Map<Param<?>, Object> paramMap = new HashMap<>();
+ private static final HashFunction HASH = murmur3_32(0);
+
+ public FeatureHasher() {
+ ParamUtils.initializeMapWithDefaultValues(paramMap, this);
+ }
+
+ @Override
+ public Table[] transform(Table... inputs) {
+ Preconditions.checkArgument(inputs.length == 1);
+ StreamTableEnvironment tEnv =
+ (StreamTableEnvironment) ((TableImpl) inputs[0]).getTableEnvironment();
+ RowTypeInfo inputTypeInfo = TableUtils.getRowTypeInfo(inputs[0].getResolvedSchema());
+ RowTypeInfo outputTypeInfo =
+ new RowTypeInfo(
+ ArrayUtils.addAll(inputTypeInfo.getFieldTypes(), VectorTypeInfo.INSTANCE),
+ ArrayUtils.addAll(inputTypeInfo.getFieldNames(), getOutputCol()));
+ DataStream<Row> output =
+ tEnv.toDataStream(inputs[0])
+ .map(
+ new HashFunc(
+ getInputCols(), getCategoricalCols(), getNumFeatures()),
+ outputTypeInfo);
+ Table outputTable = tEnv.fromDataStream(output);
+ return new Table[] {outputTable};
+ }
+
+ private static class HashFunc implements MapFunction<Row, Row> {
+ private final String[] categoricalCols;
+ private final int numFeatures;
+ private final String[] numericCols;
+
+ public HashFunc(String[] inputCols, String[] handleInvalid, int numFeatures) {
+ this.categoricalCols = handleInvalid;
+ this.numFeatures = numFeatures;
+ this.numericCols = ArrayUtils.removeElements(inputCols, categoricalCols);
+ }
+
+ @Override
+ public Row map(Row row) {
+ TreeMap<Integer, Double> feature = new TreeMap<>();
+ for (String col : numericCols) {
+ if (null != row.getField(col)) {
+ double value = ((Number) row.getFieldAs(col)).doubleValue();
+ updateMap(col, value, feature, numFeatures);
+ }
+ }
+ for (String col : categoricalCols) {
+ if (null != row.getField(col)) {
+ updateMap(col + "=" + row.getField(col), 1.0, feature, numFeatures);
+ }
+ }
+ return Row.join(row, Row.of(new SparseVector(numFeatures, feature)));
Review Comment:
It might be better to extract the indices and values from the TreeMap here and use the original constructor of `SparseVector`. I'm not sure the newly-introduced SparseVector that accepts a TreeMap would be widely used.
--
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-unsubscribe@flink.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org