msokolov commented on code in PR #12169: URL: https://github.com/apache/lucene/pull/12169#discussion_r1144077473
########## lucene/analysis/common/src/java/org/apache/lucene/analysis/synonym/word2vec/Dl4jModelReader.java: ########## @@ -0,0 +1,122 @@ +/* + * 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.lucene.analysis.synonym.word2vec; + +import java.io.BufferedInputStream; +import java.io.BufferedReader; +import java.io.IOException; +import java.io.InputStream; +import java.io.InputStreamReader; +import java.io.UnsupportedEncodingException; +import java.nio.charset.StandardCharsets; +import java.util.Base64; +import java.util.Locale; +import java.util.zip.ZipEntry; +import java.util.zip.ZipInputStream; +import org.apache.lucene.util.BytesRef; +import org.apache.lucene.util.TermAndVector; + +/** + * Word2VecModelReader is a Word2VecModelReader that reads the file generated by the library Review Comment: Should this say "Dl4jModelReader is a Word2Vec..." ? ########## lucene/analysis/common/src/java/org/apache/lucene/analysis/synonym/word2vec/Dl4jModelReader.java: ########## @@ -0,0 +1,122 @@ +/* + * 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.lucene.analysis.synonym.word2vec; + +import java.io.BufferedInputStream; +import java.io.BufferedReader; +import java.io.IOException; +import java.io.InputStream; +import java.io.InputStreamReader; +import java.io.UnsupportedEncodingException; +import java.nio.charset.StandardCharsets; +import java.util.Base64; +import java.util.Locale; +import java.util.zip.ZipEntry; +import java.util.zip.ZipInputStream; +import org.apache.lucene.util.BytesRef; +import org.apache.lucene.util.TermAndVector; + +/** + * Word2VecModelReader is a Word2VecModelReader that reads the file generated by the library + * Deeplearning4j + * + * <p>Dl4j Word2Vec documentation: + * https://deeplearning4j.konduit.ai/v/en-1.0.0-beta7/language-processing/word2vec Example to + * generate a model using dl4j: + * https://github.com/eclipse/deeplearning4j-examples/blob/master/dl4j-examples/src/main/java/org/deeplearning4j/examples/advanced/modelling/embeddingsfromcorpus/word2vec/Word2VecRawTextExample.java + * + * @lucene.experimental + */ +public class Dl4jModelReader implements Word2VecModelReader { + + private static final String MODEL_FILE_NAME_PREFIX = "syn0"; + + private final String word2vecModelFilePath; + private final ZipInputStream word2VecModelZipFile; + + public Dl4jModelReader(String word2vecModelFilePath, InputStream stream) { + this.word2vecModelFilePath = word2vecModelFilePath; + this.word2VecModelZipFile = new ZipInputStream(new BufferedInputStream(stream)); + } + + @Override + public Word2VecModel read() throws IOException { + + ZipEntry entry; + while ((entry = word2VecModelZipFile.getNextEntry()) != null) { + String fileName = entry.getName(); + if (fileName.startsWith(MODEL_FILE_NAME_PREFIX)) { + BufferedReader reader = + new BufferedReader(new InputStreamReader(word2VecModelZipFile, StandardCharsets.UTF_8)); + + String header = reader.readLine(); + String[] headerValues = header.split(" "); + int dictionarySize = Integer.parseInt(headerValues[0]); + int vectorDimension = Integer.parseInt(headerValues[1]); + + Word2VecModel model = new Word2VecModel(dictionarySize, vectorDimension); + reader + .lines() + .forEach( + line -> { + String[] tokens = line.split(" "); + BytesRef term = decodeTerm(tokens[0]); + + float[] vector = new float[tokens.length - 1]; + + if (vectorDimension != vector.length) { + throw new RuntimeException( + String.format( + Locale.ROOT, + "Word2Vec model file corrupted. " + + "Declared vectors of size %d but found vector of size %d for word %s (%s)", + vectorDimension, + vector.length, + tokens[0], + term.utf8ToString())); + } + + for (int i = 1; i < tokens.length; i++) { + vector[i - 1] = Float.parseFloat(tokens[i]); + } + model.addTermAndVector(new TermAndVector(term, vector)); + }); + return model; + } + } + throw new UnsupportedEncodingException( + "The ZIP file '" + + word2vecModelFilePath + + "' does not contain any " + + MODEL_FILE_NAME_PREFIX + + " file"); + } + + static BytesRef decodeTerm(String term) { + if (term.toLowerCase(Locale.ROOT).startsWith("b64:")) { Review Comment: It seems wasteful to lower case every term here, even when they are not b64-encoded. Also: is this something that dl4j will do consistently throughout the file? If so, we can peek at the first term and then assume the remainder will also be b64-encoded. I also wonder about the `trim()` - why do we need it? Does `Base64.decode` leave garbage at the end of the terms sometimes? ########## lucene/analysis/common/src/java/org/apache/lucene/analysis/synonym/word2vec/Word2VecModel.java: ########## @@ -0,0 +1,93 @@ +/* + * 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.lucene.analysis.synonym.word2vec; + +import java.io.IOException; +import java.util.HashMap; +import java.util.Map; +import org.apache.lucene.util.BytesRef; +import org.apache.lucene.util.TermAndVector; +import org.apache.lucene.util.hnsw.RandomAccessVectorValues; + +/** + * Word2VecModel is a class representing the parsed Word2Vec model containing the vectors for each + * word in dictionary + * + * @lucene.experimental + */ +public class Word2VecModel implements RandomAccessVectorValues<float[]> { + + private final int dictionarySize; + private final int vectorDimension; + private final TermAndVector[] data; + private final Map<BytesRef, TermAndVector> word2Vec; Review Comment: did you consider using BytesRefHash? ########## lucene/analysis/common/src/java/org/apache/lucene/analysis/synonym/word2vec/Dl4jModelReader.java: ########## @@ -0,0 +1,122 @@ +/* + * 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.lucene.analysis.synonym.word2vec; + +import java.io.BufferedInputStream; +import java.io.BufferedReader; +import java.io.IOException; +import java.io.InputStream; +import java.io.InputStreamReader; +import java.io.UnsupportedEncodingException; +import java.nio.charset.StandardCharsets; +import java.util.Base64; +import java.util.Locale; +import java.util.zip.ZipEntry; +import java.util.zip.ZipInputStream; +import org.apache.lucene.util.BytesRef; +import org.apache.lucene.util.TermAndVector; + +/** + * Word2VecModelReader is a Word2VecModelReader that reads the file generated by the library + * Deeplearning4j + * + * <p>Dl4j Word2Vec documentation: + * https://deeplearning4j.konduit.ai/v/en-1.0.0-beta7/language-processing/word2vec Example to + * generate a model using dl4j: + * https://github.com/eclipse/deeplearning4j-examples/blob/master/dl4j-examples/src/main/java/org/deeplearning4j/examples/advanced/modelling/embeddingsfromcorpus/word2vec/Word2VecRawTextExample.java + * + * @lucene.experimental + */ +public class Dl4jModelReader implements Word2VecModelReader { + + private static final String MODEL_FILE_NAME_PREFIX = "syn0"; + + private final String word2vecModelFilePath; + private final ZipInputStream word2VecModelZipFile; + + public Dl4jModelReader(String word2vecModelFilePath, InputStream stream) { + this.word2vecModelFilePath = word2vecModelFilePath; + this.word2VecModelZipFile = new ZipInputStream(new BufferedInputStream(stream)); + } + + @Override + public Word2VecModel read() throws IOException { + + ZipEntry entry; + while ((entry = word2VecModelZipFile.getNextEntry()) != null) { + String fileName = entry.getName(); + if (fileName.startsWith(MODEL_FILE_NAME_PREFIX)) { + BufferedReader reader = + new BufferedReader(new InputStreamReader(word2VecModelZipFile, StandardCharsets.UTF_8)); + + String header = reader.readLine(); + String[] headerValues = header.split(" "); + int dictionarySize = Integer.parseInt(headerValues[0]); + int vectorDimension = Integer.parseInt(headerValues[1]); + + Word2VecModel model = new Word2VecModel(dictionarySize, vectorDimension); + reader + .lines() + .forEach( + line -> { + String[] tokens = line.split(" "); + BytesRef term = decodeTerm(tokens[0]); + + float[] vector = new float[tokens.length - 1]; + + if (vectorDimension != vector.length) { + throw new RuntimeException( + String.format( + Locale.ROOT, + "Word2Vec model file corrupted. " + + "Declared vectors of size %d but found vector of size %d for word %s (%s)", + vectorDimension, + vector.length, + tokens[0], + term.utf8ToString())); + } + + for (int i = 1; i < tokens.length; i++) { + vector[i - 1] = Float.parseFloat(tokens[i]); + } + model.addTermAndVector(new TermAndVector(term, vector)); + }); + return model; + } + } + throw new UnsupportedEncodingException( Review Comment: I think this exception is really intended for use with character set encodings only. Maybe IllegalArgumentException would fit better? ########## lucene/analysis/common/src/java/org/apache/lucene/analysis/synonym/word2vec/SynonymProvider.java: ########## @@ -0,0 +1,42 @@ +/* + * 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.lucene.analysis.synonym.word2vec; + +import java.io.IOException; +import java.util.List; +import org.apache.lucene.util.BytesRef; +import org.apache.lucene.util.TermAndBoost; + +/** + * Generic synonym provider + * + * @lucene.experimental + */ +public interface SynonymProvider { + + /** + * SynonymProvider constructor + * + * @param term we want to find the synonyms + * @param maxSynonymsPerTerm maximum number of result returned by the synonym search Review Comment: I don't see that we need this interface if its only use is in this one `Dl4jWord2VecSynonymFilter`. Can't we simply refer directly to the implementing class? ########## lucene/analysis/common/src/java/org/apache/lucene/analysis/synonym/word2vec/Dl4jModelReader.java: ########## @@ -0,0 +1,122 @@ +/* + * 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.lucene.analysis.synonym.word2vec; + +import java.io.BufferedInputStream; +import java.io.BufferedReader; +import java.io.IOException; +import java.io.InputStream; +import java.io.InputStreamReader; +import java.io.UnsupportedEncodingException; +import java.nio.charset.StandardCharsets; +import java.util.Base64; +import java.util.Locale; +import java.util.zip.ZipEntry; +import java.util.zip.ZipInputStream; +import org.apache.lucene.util.BytesRef; +import org.apache.lucene.util.TermAndVector; + +/** + * Word2VecModelReader is a Word2VecModelReader that reads the file generated by the library + * Deeplearning4j + * + * <p>Dl4j Word2Vec documentation: + * https://deeplearning4j.konduit.ai/v/en-1.0.0-beta7/language-processing/word2vec Example to + * generate a model using dl4j: + * https://github.com/eclipse/deeplearning4j-examples/blob/master/dl4j-examples/src/main/java/org/deeplearning4j/examples/advanced/modelling/embeddingsfromcorpus/word2vec/Word2VecRawTextExample.java + * + * @lucene.experimental + */ +public class Dl4jModelReader implements Word2VecModelReader { + + private static final String MODEL_FILE_NAME_PREFIX = "syn0"; + + private final String word2vecModelFilePath; Review Comment: why pass in the path when it is only used in `toString()`? Can we choose between accepting a `java.nio.Path` that does its own open/close of the zip file and an (anonymous) `InputStream`? ########## lucene/analysis/common/src/java/org/apache/lucene/analysis/synonym/word2vec/Word2VecModelReader.java: ########## @@ -0,0 +1,32 @@ +/* + * 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.lucene.analysis.synonym.word2vec; + +import java.io.Closeable; +import java.io.IOException; + +/** + * Each class extending this interface must be able to read a Word2Vec model format and provide a + * Word2VecModel with normalized vectors + * + * @lucene.experimental + */ +public interface Word2VecModelReader extends Closeable { Review Comment: Do we have more than one implementation? I don't think this interface is necessary. Later we can always add it if we have multiple implementations and need to abstract. For now it's just extra stuff to maintain ########## lucene/analysis/common/src/java/org/apache/lucene/analysis/synonym/word2vec/Dl4jModelReader.java: ########## @@ -0,0 +1,122 @@ +/* + * 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.lucene.analysis.synonym.word2vec; + +import java.io.BufferedInputStream; +import java.io.BufferedReader; +import java.io.IOException; +import java.io.InputStream; +import java.io.InputStreamReader; +import java.io.UnsupportedEncodingException; +import java.nio.charset.StandardCharsets; +import java.util.Base64; +import java.util.Locale; +import java.util.zip.ZipEntry; +import java.util.zip.ZipInputStream; +import org.apache.lucene.util.BytesRef; +import org.apache.lucene.util.TermAndVector; + +/** + * Word2VecModelReader is a Word2VecModelReader that reads the file generated by the library + * Deeplearning4j + * + * <p>Dl4j Word2Vec documentation: + * https://deeplearning4j.konduit.ai/v/en-1.0.0-beta7/language-processing/word2vec Example to + * generate a model using dl4j: + * https://github.com/eclipse/deeplearning4j-examples/blob/master/dl4j-examples/src/main/java/org/deeplearning4j/examples/advanced/modelling/embeddingsfromcorpus/word2vec/Word2VecRawTextExample.java + * + * @lucene.experimental + */ +public class Dl4jModelReader implements Word2VecModelReader { + + private static final String MODEL_FILE_NAME_PREFIX = "syn0"; + + private final String word2vecModelFilePath; + private final ZipInputStream word2VecModelZipFile; + + public Dl4jModelReader(String word2vecModelFilePath, InputStream stream) { + this.word2vecModelFilePath = word2vecModelFilePath; + this.word2VecModelZipFile = new ZipInputStream(new BufferedInputStream(stream)); + } + + @Override + public Word2VecModel read() throws IOException { + + ZipEntry entry; + while ((entry = word2VecModelZipFile.getNextEntry()) != null) { + String fileName = entry.getName(); + if (fileName.startsWith(MODEL_FILE_NAME_PREFIX)) { + BufferedReader reader = + new BufferedReader(new InputStreamReader(word2VecModelZipFile, StandardCharsets.UTF_8)); + + String header = reader.readLine(); + String[] headerValues = header.split(" "); + int dictionarySize = Integer.parseInt(headerValues[0]); + int vectorDimension = Integer.parseInt(headerValues[1]); + + Word2VecModel model = new Word2VecModel(dictionarySize, vectorDimension); + reader Review Comment: to me this would read much clearer using a traditional `while` loop: ``` while ((line = reader.readLine()) != null) ... ``` ########## lucene/analysis/common/src/java/org/apache/lucene/analysis/synonym/word2vec/Word2VecSynonymFilter.java: ########## @@ -0,0 +1,111 @@ +/* + * 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.lucene.analysis.synonym.word2vec; + +import java.io.IOException; +import java.util.LinkedList; +import java.util.List; +import org.apache.lucene.analysis.TokenFilter; +import org.apache.lucene.analysis.TokenStream; +import org.apache.lucene.analysis.synonym.SynonymGraphFilter; +import org.apache.lucene.analysis.tokenattributes.CharTermAttribute; +import org.apache.lucene.analysis.tokenattributes.PositionIncrementAttribute; +import org.apache.lucene.analysis.tokenattributes.PositionLengthAttribute; +import org.apache.lucene.analysis.tokenattributes.TypeAttribute; +import org.apache.lucene.search.BoostAttribute; +import org.apache.lucene.util.BytesRef; +import org.apache.lucene.util.BytesRefBuilder; +import org.apache.lucene.util.TermAndBoost; + +/** + * Applies single-token synonyms from a Word2Vec trained network to an incoming {@link TokenStream}. Review Comment: I don't see any reference to word2vec in here? We just use this abstract SynonymFilter that could conceivably be implemented using a thesaurus. That makes this more powerful than I think we are claiming it to be. Again, let's simply use the concrete classes throughout and call the classes `Dl4jWord2VecXXX` ########## lucene/analysis/common/src/java/org/apache/lucene/analysis/synonym/word2vec/Word2VecModel.java: ########## @@ -0,0 +1,93 @@ +/* + * 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.lucene.analysis.synonym.word2vec; + +import java.io.IOException; +import java.util.HashMap; +import java.util.Map; +import org.apache.lucene.util.BytesRef; +import org.apache.lucene.util.TermAndVector; +import org.apache.lucene.util.hnsw.RandomAccessVectorValues; + +/** + * Word2VecModel is a class representing the parsed Word2Vec model containing the vectors for each + * word in dictionary + * + * @lucene.experimental + */ +public class Word2VecModel implements RandomAccessVectorValues<float[]> { + + private final int dictionarySize; + private final int vectorDimension; + private final TermAndVector[] data; + private final Map<BytesRef, TermAndVector> word2Vec; + private int loadedCount = 0; + + public Word2VecModel(int dictionarySize, int vectorDimension) { + this.dictionarySize = dictionarySize; + this.vectorDimension = vectorDimension; + this.data = new TermAndVector[dictionarySize]; + this.word2Vec = new HashMap<>(); + } + + private Word2VecModel( + int dictionarySize, + int vectorDimension, + TermAndVector[] data, + Map<BytesRef, TermAndVector> word2Vec) { + this.dictionarySize = dictionarySize; + this.vectorDimension = vectorDimension; + this.data = data; + this.word2Vec = word2Vec; + } + + public void addTermAndVector(TermAndVector modelEntry) { + modelEntry.normalizeVector(); + this.data[loadedCount++] = modelEntry; + this.word2Vec.put(modelEntry.getTerm(), modelEntry); + } + + @Override + public float[] vectorValue(int ord) throws IOException { + return data[ord].getVector(); + } + + public float[] vectorValue(BytesRef term) { + TermAndVector entry = word2Vec.get(term); + return (entry == null) ? null : entry.getVector(); + } + + public BytesRef binaryValue(int targetOrd) throws IOException { + return data[targetOrd].getTerm(); + } + + @Override + public int dimension() { + return vectorDimension; + } + + @Override + public int size() { + return dictionarySize; + } + + @Override + public RandomAccessVectorValues<float[]> copy() throws IOException { Review Comment: It does not need to do a deep copy; the purpose of this method is to enable multiple concurrent accesses to the underlying data. Since this implementation doesn't have any temporary variable into which vectors are decoded (which could be overwritten), I think it's safe to simply `return this`. ########## lucene/analysis/common/src/java/org/apache/lucene/analysis/synonym/word2vec/Word2VecModel.java: ########## @@ -0,0 +1,93 @@ +/* + * 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.lucene.analysis.synonym.word2vec; + +import java.io.IOException; +import java.util.HashMap; +import java.util.Map; +import org.apache.lucene.util.BytesRef; +import org.apache.lucene.util.TermAndVector; +import org.apache.lucene.util.hnsw.RandomAccessVectorValues; + +/** + * Word2VecModel is a class representing the parsed Word2Vec model containing the vectors for each + * word in dictionary + * + * @lucene.experimental + */ +public class Word2VecModel implements RandomAccessVectorValues<float[]> { + + private final int dictionarySize; + private final int vectorDimension; + private final TermAndVector[] data; + private final Map<BytesRef, TermAndVector> word2Vec; + private int loadedCount = 0; + + public Word2VecModel(int dictionarySize, int vectorDimension) { + this.dictionarySize = dictionarySize; + this.vectorDimension = vectorDimension; + this.data = new TermAndVector[dictionarySize]; + this.word2Vec = new HashMap<>(); + } + + private Word2VecModel( + int dictionarySize, + int vectorDimension, + TermAndVector[] data, + Map<BytesRef, TermAndVector> word2Vec) { + this.dictionarySize = dictionarySize; + this.vectorDimension = vectorDimension; + this.data = data; + this.word2Vec = word2Vec; + } + + public void addTermAndVector(TermAndVector modelEntry) { + modelEntry.normalizeVector(); + this.data[loadedCount++] = modelEntry; + this.word2Vec.put(modelEntry.getTerm(), modelEntry); + } + + @Override + public float[] vectorValue(int ord) throws IOException { + return data[ord].getVector(); + } + + public float[] vectorValue(BytesRef term) { + TermAndVector entry = word2Vec.get(term); + return (entry == null) ? null : entry.getVector(); + } + + public BytesRef binaryValue(int targetOrd) throws IOException { + return data[targetOrd].getTerm(); Review Comment: This is incorrect - the purpose of this method is to return some bytes representing the vector value. I think instead you ought to simply throw `UnsupportedOperationException` since this implementation will never be used in an indexing context where this method is required. ########## lucene/analysis/common/src/java/org/apache/lucene/analysis/synonym/word2vec/Word2VecSynonymFilter.java: ########## @@ -0,0 +1,111 @@ +/* + * 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.lucene.analysis.synonym.word2vec; + +import java.io.IOException; +import java.util.LinkedList; +import java.util.List; +import org.apache.lucene.analysis.TokenFilter; +import org.apache.lucene.analysis.TokenStream; +import org.apache.lucene.analysis.synonym.SynonymGraphFilter; +import org.apache.lucene.analysis.tokenattributes.CharTermAttribute; +import org.apache.lucene.analysis.tokenattributes.PositionIncrementAttribute; +import org.apache.lucene.analysis.tokenattributes.PositionLengthAttribute; +import org.apache.lucene.analysis.tokenattributes.TypeAttribute; +import org.apache.lucene.search.BoostAttribute; +import org.apache.lucene.util.BytesRef; +import org.apache.lucene.util.BytesRefBuilder; +import org.apache.lucene.util.TermAndBoost; + +/** + * Applies single-token synonyms from a Word2Vec trained network to an incoming {@link TokenStream}. + * + * @lucene.experimental + */ +public final class Word2VecSynonymFilter extends TokenFilter { + + private final CharTermAttribute termAtt = addAttribute(CharTermAttribute.class); + private final PositionIncrementAttribute posIncrementAtt = + addAttribute(PositionIncrementAttribute.class); + private final PositionLengthAttribute posLenAtt = addAttribute(PositionLengthAttribute.class); + private final BoostAttribute boostAtt = addAttribute(BoostAttribute.class); + private final TypeAttribute typeAtt = addAttribute(TypeAttribute.class); + + private final SynonymProvider synonymProvider; + private final int maxSynonymsPerTerm; + private final float minAcceptedSimilarity; + private final LinkedList<TermAndBoost> synonymBuffer = new LinkedList<>(); + private State lastState; + + /** + * Apply previously built synonymProvider to incoming tokens. + * + * @param input input tokenstream + * @param synonymProvider synonym provider + * @param maxSynonymsPerTerm maximum number of result returned by the synonym search + * @param minAcceptedSimilarity minimal value of cosine similarity between the searched vector and + * the retrieved ones + */ + public Word2VecSynonymFilter( + TokenStream input, + SynonymProvider synonymProvider, + int maxSynonymsPerTerm, + float minAcceptedSimilarity) { + super(input); + this.synonymProvider = synonymProvider; + this.maxSynonymsPerTerm = maxSynonymsPerTerm; + this.minAcceptedSimilarity = minAcceptedSimilarity; + } + + @Override + public boolean incrementToken() throws IOException { + + if (!synonymBuffer.isEmpty()) { + TermAndBoost synonym = synonymBuffer.pollFirst(); + clearAttributes(); + restoreState(this.lastState); + termAtt.setEmpty(); Review Comment: this would be cleaner if we used `BytesTermAttribute` ########## lucene/analysis/common/src/java/org/apache/lucene/analysis/synonym/word2vec/Dl4jModelReader.java: ########## @@ -0,0 +1,122 @@ +/* + * 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.lucene.analysis.synonym.word2vec; + +import java.io.BufferedInputStream; +import java.io.BufferedReader; +import java.io.IOException; +import java.io.InputStream; +import java.io.InputStreamReader; +import java.io.UnsupportedEncodingException; +import java.nio.charset.StandardCharsets; +import java.util.Base64; +import java.util.Locale; +import java.util.zip.ZipEntry; +import java.util.zip.ZipInputStream; +import org.apache.lucene.util.BytesRef; +import org.apache.lucene.util.TermAndVector; + +/** + * Word2VecModelReader is a Word2VecModelReader that reads the file generated by the library + * Deeplearning4j + * + * <p>Dl4j Word2Vec documentation: + * https://deeplearning4j.konduit.ai/v/en-1.0.0-beta7/language-processing/word2vec Example to + * generate a model using dl4j: + * https://github.com/eclipse/deeplearning4j-examples/blob/master/dl4j-examples/src/main/java/org/deeplearning4j/examples/advanced/modelling/embeddingsfromcorpus/word2vec/Word2VecRawTextExample.java + * + * @lucene.experimental + */ +public class Dl4jModelReader implements Word2VecModelReader { Review Comment: +1 to avoid dependencies whenever practical -- 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...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
