jiangxin369 commented on code in PR #191: URL: https://github.com/apache/flink-ml/pull/191#discussion_r1057228565
########## flink-ml-lib/src/main/java/org/apache/flink/ml/feature/lsh/MinHashLSH.java: ########## @@ -0,0 +1,57 @@ +/* + * 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.lsh; + +import org.apache.flink.ml.util.ReadWriteUtils; +import org.apache.flink.streaming.api.datastream.DataStream; +import org.apache.flink.table.api.bridge.java.StreamTableEnvironment; + +import java.io.IOException; + +/** + * An Estimator that implements the MinHash LSH algorithm, which supports LSH for Jaccard distance. + * + * <p>The input could be dense or sparse vectors. Each input vector must hava at least one non-zero Review Comment: ```suggestion * <p>The input could be dense or sparse vectors. Each input vector must have at least one non-zero ``` ########## flink-ml-lib/src/main/java/org/apache/flink/ml/feature/lsh/LSHModel.java: ########## @@ -0,0 +1,457 @@ +/* + * 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.lsh; + +import org.apache.flink.api.common.functions.AggregateFunction; +import org.apache.flink.api.common.functions.FlatMapFunction; +import org.apache.flink.api.common.functions.RichFlatMapFunction; +import org.apache.flink.api.common.functions.RichMapFunction; +import org.apache.flink.api.common.typeinfo.TypeInformation; +import org.apache.flink.api.common.typeinfo.Types; +import org.apache.flink.api.java.functions.KeySelector; +import org.apache.flink.api.java.tuple.Tuple2; +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.DataStreamUtils; +import org.apache.flink.ml.common.datastream.EndOfStreamWindows; +import org.apache.flink.ml.common.datastream.TableUtils; +import org.apache.flink.ml.linalg.DenseVector; +import org.apache.flink.ml.linalg.Vector; +import org.apache.flink.ml.param.Param; +import org.apache.flink.ml.util.ParamUtils; +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.Collector; +import org.apache.flink.util.Preconditions; + +import org.apache.commons.lang3.ArrayUtils; + +import java.util.ArrayList; +import java.util.Collections; +import java.util.Comparator; +import java.util.HashMap; +import java.util.List; +import java.util.Map; +import java.util.PriorityQueue; + +/** + * Base class for LSH model. + * + * <p>In addition to transforming input feature vectors to multiple hash values, it also supports + * approximate nearest neighbors search within a dataset regarding a key vector and approximate + * similarity join between two datasets. + * + * @param <T> class type of the LSHModel implementation itself. + */ +abstract class LSHModel<T extends LSHModel<T>> implements Model<T>, LSHModelParams<T> { + private static final String MODEL_DATA_BC_KEY = "modelData"; + + private final Map<Param<?>, Object> paramMap = new HashMap<>(); + + /** Stores the corresponding model data class of T. */ + private final Class<? extends LSHModelData> modelDataClass; + + protected Table modelDataTable; + + public LSHModel(Class<? extends LSHModelData> modelDataClass) { + this.modelDataClass = modelDataClass; + ParamUtils.initializeMapWithDefaultValues(paramMap, this); + } + + @Override + public T setModelData(Table... inputs) { + modelDataTable = inputs[0]; + return (T) this; + } + + @Override + public Table[] getModelData() { + return new Table[] {modelDataTable}; + } + + @Override + public Map<Param<?>, Object> getParamMap() { + return paramMap; + } + + @Override + public Table[] transform(Table... inputs) { + Preconditions.checkArgument(inputs.length == 1); + StreamTableEnvironment tEnv = + (StreamTableEnvironment) ((TableImpl) inputs[0]).getTableEnvironment(); + DataStream<? extends LSHModelData> modelData = + tEnv.toDataStream(modelDataTable, modelDataClass); + + RowTypeInfo inputTypeInfo = TableUtils.getRowTypeInfo(inputs[0].getResolvedSchema()); + TypeInformation<?> outputType = TypeInformation.of(DenseVector[].class); + RowTypeInfo outputTypeInfo = + new RowTypeInfo( + ArrayUtils.addAll(inputTypeInfo.getFieldTypes(), outputType), + ArrayUtils.addAll(inputTypeInfo.getFieldNames(), getOutputCol())); + + DataStream<Row> output = + BroadcastUtils.withBroadcastStream( + Collections.singletonList(tEnv.toDataStream(inputs[0])), + Collections.singletonMap(MODEL_DATA_BC_KEY, modelData), + inputList -> { + //noinspection unchecked + DataStream<Row> data = (DataStream<Row>) inputList.get(0); + return data.map(new PredictFunction(getInputCol()), outputTypeInfo); + }); + return new Table[] {tEnv.fromDataStream(output)}; + } + + /** + * Approximately finds at most k items from a dataset which have the closest distance to a given + * item . If the `outputCol` is missing in the given dataset, this method transforms the dataset Review Comment: ```suggestion * item. If the `outputCol` is missing in the given dataset, this method transforms the dataset ``` ########## docs/content/docs/operators/feature/minhashlsh.md: ########## @@ -0,0 +1,276 @@ +--- +title: "MinHash LSH" +weight: 1 +type: docs +aliases: +- /operators/feature/minhashlsh.html +--- + +<!-- +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. +--> + +## MinHash LSH + +MinHash LSH is a Locality Sensitive Hashing (LSH) scheme for Jaccard distance metric. +The input features are sets of natural numbers represented as non-zero indices of vectors, +either dense vectors or sparse vectors. Typically, sparse vectors are more efficient. + +### Input Columns + +| Param name | Type | Default | Description | +|:-----------|:-------|:----------|:-----------------------| +| inputCol | Vector | `"input"` | Features to be mapped. | + +### Output Columns + +| Param name | Type | Default | Description | +|:-----------|:--------------|:-----------|:-------------| +| outputCol | DenseVector[] | `"output"` | Hash values. | + +### Parameters + +| Key | Default | Type | Required | Description | +|-------------------------|-----------------------------------------------------------|---------|----------|--------------------------------------------------------------------| +| inputCol | `"input"` | String | no | Input column name. | +| outputCol | `"output"` | String | no | Output column name. | +| seed | `"org.apache.flink.ml.feature.lsh.MinHashLSH".hashCode()` | Long | no | The random seed. | Review Comment: I think `null` is ok for the default value of `seed`, just like in `KMeans` doc. ########## flink-ml-python/pyflink/ml/lib/feature/lsh.py: ########## @@ -0,0 +1,191 @@ +################################################################################ +# 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. +################################################################################ + +import typing +from abc import ABC +from pyflink.java_gateway import get_gateway +from pyflink.table import Table +from pyflink.util.java_utils import to_jarray + +from pyflink.ml.core.linalg import Vector, DenseVector, SparseVector +from pyflink.ml.core.param import Param, IntParam, ParamValidators +from pyflink.ml.core.wrapper import JavaWithParams +from pyflink.ml.lib.feature.common import JavaFeatureEstimator, JavaFeatureModel +from pyflink.ml.lib.param import HasInputCol, HasOutputCol, HasSeed + + +class _LSHParams( + JavaWithParams, + HasInputCol, + HasOutputCol, + HasSeed +): + """ + Params for :class:`LSH` + """ + + NUM_HASH_TABLES: Param[int] = IntParam( + "num_hash_tables", "Number of hash tables.", 1, ParamValidators.gt_eq(1) + ) + + NUM_HASH_FUNCTIONS_PER_TABLE: Param[int] = IntParam( + "num_hash_functions_per_table", + "Number of hash functions per table.", + 1, + ParamValidators.gt_eq(1.)) + + def __init__(self, java_params): + super(_LSHParams, self).__init__(java_params) + + def set_num_hash_tables(self, value: int): + return typing.cast(_LSHParams, self.set(self.NUM_HASH_TABLES, value)) + + def get_num_hash_tables(self): + return self.get(self.NUM_HASH_TABLES) + + @property + def num_hash_tables(self): + return self.get_num_hash_tables() + + def set_num_hash_functions_per_table(self, value: int): + return typing.cast(_LSHParams, self.set(self.NUM_HASH_FUNCTIONS_PER_TABLE, value)) + + def get_num_hash_functions_per_table(self): + return self.get(self.NUM_HASH_FUNCTIONS_PER_TABLE) + + @property + def num_hash_functions_per_table(self): + return self.get_num_hash_functions_per_table() + + +class _LSHModelParams(JavaWithParams, + HasInputCol, + HasOutputCol): + ... Review Comment: Why it has no `__init__` function? ########## docs/content/docs/operators/feature/minhashlsh.md: ########## @@ -0,0 +1,276 @@ +--- +title: "MinHash LSH" +weight: 1 +type: docs +aliases: +- /operators/feature/minhashlsh.html +--- + +<!-- +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. +--> + +## MinHash LSH + +MinHash LSH is a Locality Sensitive Hashing (LSH) scheme for Jaccard distance metric. +The input features are sets of natural numbers represented as non-zero indices of vectors, +either dense vectors or sparse vectors. Typically, sparse vectors are more efficient. Review Comment: Not like the other algorithms, LSH has additional public APIs like `Approximate Similarity Join` and `Approximate Nearest Neighbor Search`, do you think we need to add documents for them? ########## flink-ml-lib/src/main/java/org/apache/flink/ml/feature/lsh/LSH.java: ########## @@ -0,0 +1,104 @@ +/* + * 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.lsh; + +import org.apache.flink.ml.api.Estimator; +import org.apache.flink.ml.common.datastream.DataStreamUtils; +import org.apache.flink.ml.linalg.Vector; +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 java.io.IOException; +import java.util.HashMap; +import java.util.Map; + +/** + * Base class for estimators that support LSH (Locality-sensitive hashing) algorithm for different + * metrics (e.g., Jaccard distance). + * + * <p>The basic idea of LSH is to use to a set of hash functions to map input vectors into different Review Comment: ```suggestion * <p>The basic idea of LSH is to use a set of hash functions to map input vectors into different ``` ########## flink-ml-lib/src/main/java/org/apache/flink/ml/feature/lsh/LSHParams.java: ########## @@ -0,0 +1,74 @@ +/* + * 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.lsh; + +import org.apache.flink.ml.common.param.HasInputCol; +import org.apache.flink.ml.common.param.HasOutputCol; +import org.apache.flink.ml.common.param.HasSeed; +import org.apache.flink.ml.param.IntParam; +import org.apache.flink.ml.param.Param; +import org.apache.flink.ml.param.ParamValidators; + +/** + * Params for {@link LSH}. + * + * @param <T> The class type of this instance. + */ +public interface LSHParams<T> extends HasInputCol<T>, HasOutputCol<T>, HasSeed<T> { Review Comment: Seems that the param `seed` is only used in `MinHashLSH`, how about moving the `HasSeed` to `MinHashLSH`? Besides, how about extending `LSHModelParams<T>` instead of `HasInputCol<T>, HasOutputCol<T>`, just like the other algorithms? ########## flink-ml-lib/src/main/java/org/apache/flink/ml/feature/lsh/LSH.java: ########## @@ -0,0 +1,104 @@ +/* + * 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.lsh; + +import org.apache.flink.ml.api.Estimator; +import org.apache.flink.ml.common.datastream.DataStreamUtils; +import org.apache.flink.ml.linalg.Vector; +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 java.io.IOException; +import java.util.HashMap; +import java.util.Map; + +/** + * Base class for estimators that support LSH (Locality-sensitive hashing) algorithm for different + * metrics (e.g., Jaccard distance). + * + * <p>The basic idea of LSH is to use to a set of hash functions to map input vectors into different + * buckets, where closer vectors are expected to be in same bucket with higher probabilities. In Review Comment: ```suggestion * buckets, where closer vectors are expected to be in the same bucket with higher probabilities. In ``` -- 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]
