use of edu.stanford.nlp.parser.dvparser.DVModelReranker in project CoreNLP by stanfordnlp.
the class ConvertModels method readParser.
public static LexicalizedParser readParser(ObjectInputStream in) throws IOException, ClassNotFoundException {
LexicalizedParser model = ErasureUtils.uncheckedCast(in.readObject());
Function<List<List<Double>>, SimpleMatrix> f = (x) -> toMatrix(x);
TwoDimensionalMap<String, String, List<List<Double>>> map2dSM = ErasureUtils.uncheckedCast(in.readObject());
TwoDimensionalMap<String, String, SimpleMatrix> binaryTransform = transform2DMap(map2dSM, f);
Map<String, List<List<Double>>> map = ErasureUtils.uncheckedCast(in.readObject());
Map<String, SimpleMatrix> unaryTransform = transformMap(map, f);
map2dSM = ErasureUtils.uncheckedCast(in.readObject());
TwoDimensionalMap<String, String, SimpleMatrix> binaryScore = transform2DMap(map2dSM, f);
map = ErasureUtils.uncheckedCast(in.readObject());
Map<String, SimpleMatrix> unaryScore = transformMap(map, f);
map = ErasureUtils.uncheckedCast(in.readObject());
Map<String, SimpleMatrix> wordVectors = transformMap(map, f);
DVModel dvModel = new DVModel(binaryTransform, unaryTransform, binaryScore, unaryScore, wordVectors, model.getOp());
DVModelReranker reranker = new DVModelReranker(dvModel);
model.reranker = reranker;
return model;
}
use of edu.stanford.nlp.parser.dvparser.DVModelReranker in project CoreNLP by stanfordnlp.
the class ConvertModels method main.
/**
* This program converts a sentiment model or an RNN parser model
* from EJML v23, used by CoreNLP 3.9.2, to a more recent version of
* EJML, such as v38. The reason for this is that the EJML v31
* update changed the serialization of SimpleMatrix in a way that
* broke all the old models, so for years we never bit the bullet of
* upgrading. This script handles the upgrade of our models which
* use SimpleMatrix. It needs to be done in two steps.
* <br>
* The first conversion turns a model into a file with lists of
* doubles in place of the SimpleMatrix objects used in ejml.
* <br>
* The second conversion turns the lists of doubles back into a new
* SentimentModel or RNN parser.
* <br>
* In between the two steps you should replace the EJML version you
* are using with 0.38, although no one is judging you if you
* aimlessly convert back and forth using the same EJML version.
* <br>
* Concrete steps:
* <br>
* <code> java edu.stanford.nlp.neural.ConvertModels -stage OLD -model SENTIMENT -input edu/stanford/nlp/models/sentiment/sentiment.ser.gz -output sentiment.INT.ser.gz</code>
* <br>
* ... update EJML library to v38 or a later
* <br>
* <code> java edu.stanford.nlp.neural.ConvertModels -stage NEW -model SENTIMENT -input sentiment.INT.ser.gz -output sentiment.38.ser.gz</code>
* <br>
* Congratulations, your old model will now work with EJML v38.
* <br>
* To upgrade an RNN model (did anyone train this themselves?) use <code>-model DVPARSER</code>
* <br>
* <code> java edu.stanford.nlp.neural.ConvertModels -stage OLD -model DVPARSER -input /u/nlp/data/lexparser/chineseRNN.e21.ser.gz -output /u/nlp/data/lexparser/chineseRNN.INT.ser.gz</code>
* <br>
* <code> java edu.stanford.nlp.neural.ConvertModels -stage NEW -model DVPARSER -input /u/nlp/data/lexparser/chineseRNN.INT.ser.gz -output /u/nlp/data/lexparser/chineseRNN.e38.ser.gz</code>
* <br>
* To upgrade a neural coref model, use <code>-model COREF</code>
* <br>
* <code> java edu.stanford.nlp.neural.ConvertModels -stage OLD -model COREF -input /scr/nlp/data/coref/models/neural/english/english-model-default.ser.gz -output /scr/nlp/data/coref/models/neural/english/english-model-default.INT.ser.gz</code>
* <br>
* <code> java edu.stanford.nlp.neural.ConvertModels -stage NEW -model COREF -input /scr/nlp/data/coref/models/neural/english/english-model-default.INT.ser.gz -output /scr/nlp/data/coref/models/neural/english/english-model-default.e39.ser.gz</code>
* <br>
* Neural coref models ship with a separate embedding file, although these are also embedded in the model itself. To upgrade this, use <code>-model EMBEDDING</code>
* <br>
* <code> java edu.stanford.nlp.neural.ConvertModels -stage OLD -model EMBEDDING -input /scr/nlp/data/coref/models/neural/english/english-embeddings.e38.ser.gz -output /scr/nlp/data/coref/models/neural/english/english-embeddings.INT.ser.gz</code>
* <br>
* <code> java edu.stanford.nlp.neural.ConvertModels -stage NEW -model EMBEDDING -input /scr/nlp/data/coref/models/neural/english/english-embeddings.INT.ser.gz -output /scr/nlp/data/coref/models/neural/english/english-embeddings.e39.ser.gz</code>
* <br>
* There is another coref model which isn't used in corenlp, but it might be in the future. To upgrade this, use <code>-model FASTCOREF</code>
* <br>
* <code> java edu.stanford.nlp.neural.ConvertModels -stage OLD -model FASTCOREF -input /scr/nlp/data/coref/models/fastneural/fast-english-model.e38.ser.gz -output /scr/nlp/data/coref/models/fastneural/fast-english-model.INT.ser.gz</code>
* <br>
* <code> java edu.stanford.nlp.neural.ConvertModels -stage NEW -model FASTCOREF -input /scr/nlp/data/coref/models/fastneural/fast-english-model.INT.ser.gz -output /scr/nlp/data/coref/models/fastneural/fast-english-model.e39.ser.gz</code>
* <br>
*
* @author <a href=horatio@gmail.com>John Bauer</a>
*/
public static void main(String[] args) throws IOException, ClassNotFoundException, InstantiationException, NoSuchMethodException {
Properties props = StringUtils.argsToProperties(args);
final Stage stage;
try {
stage = Stage.valueOf(props.getProperty("stage").toUpperCase(Locale.ROOT));
} catch (IllegalArgumentException | NullPointerException e) {
throw new IllegalArgumentException("Please specify -stage, either OLD or NEW");
}
final Model modelType;
try {
modelType = Model.valueOf(props.getProperty("model").toUpperCase(Locale.ROOT));
} catch (IllegalArgumentException | NullPointerException e) {
throw new IllegalArgumentException("Please specify -model, either SENTIMENT, DVPARSER, EMBEDDING, COREF, FASTCOREF");
}
if (!props.containsKey("input")) {
throw new IllegalArgumentException("Please specify -input");
}
if (!props.containsKey("output")) {
throw new IllegalArgumentException("Please specify -output");
}
String inputPath = props.getProperty("input");
String outputPath = props.getProperty("output");
if (modelType == Model.SENTIMENT) {
if (stage == Stage.OLD) {
SentimentModel model = SentimentModel.loadSerialized(inputPath);
ObjectOutputStream out = IOUtils.writeStreamFromString(outputPath);
writeSentiment(model, out);
out.close();
} else {
ObjectInputStream in = IOUtils.readStreamFromString(inputPath);
SentimentModel model = readSentiment(in);
in.close();
model.saveSerialized(outputPath);
}
} else if (modelType == Model.DVPARSER) {
if (stage == Stage.OLD) {
LexicalizedParser model = LexicalizedParser.loadModel(inputPath);
if (model.reranker == null) {
System.out.println("Nothing to do for " + inputPath);
} else {
// will barf if not successful
DVModelReranker reranker = (DVModelReranker) model.reranker;
model.reranker = null;
ObjectOutputStream out = IOUtils.writeStreamFromString(outputPath);
writeParser(model, reranker, out);
out.close();
}
} else {
ObjectInputStream in = IOUtils.readStreamFromString(inputPath);
LexicalizedParser model = readParser(in);
in.close();
model.saveParserToSerialized(outputPath);
}
} else if (modelType == Model.EMBEDDING) {
if (stage == Stage.OLD) {
Embedding embedding = ErasureUtils.uncheckedCast(IOUtils.readObjectFromURLOrClasspathOrFileSystem(inputPath));
ObjectOutputStream out = IOUtils.writeStreamFromString(outputPath);
writeEmbedding(embedding, out);
out.close();
} else {
ObjectInputStream in = IOUtils.readStreamFromString(inputPath);
Embedding embedding = readEmbedding(in);
in.close();
IOUtils.writeObjectToFile(embedding, outputPath);
}
} else if (modelType == Model.COREF) {
if (stage == Stage.OLD) {
NeuralCorefModel model = ErasureUtils.uncheckedCast(IOUtils.readObjectFromURLOrClasspathOrFileSystem(inputPath));
ObjectOutputStream out = IOUtils.writeStreamFromString(outputPath);
writeCoref(model, out);
out.close();
} else {
ObjectInputStream in = IOUtils.readStreamFromString(inputPath);
NeuralCorefModel model = readCoref(in);
in.close();
IOUtils.writeObjectToFile(model, outputPath);
}
} else if (modelType == Model.FASTCOREF) {
if (stage == Stage.OLD) {
FastNeuralCorefModel model = ErasureUtils.uncheckedCast(IOUtils.readObjectFromURLOrClasspathOrFileSystem(inputPath));
ObjectOutputStream out = IOUtils.writeStreamFromString(outputPath);
writeFastCoref(model, out);
out.close();
} else {
ObjectInputStream in = IOUtils.readStreamFromString(inputPath);
FastNeuralCorefModel model = readFastCoref(in);
in.close();
IOUtils.writeObjectToFile(model, outputPath);
}
} else {
throw new IllegalArgumentException("Unknown model type " + modelType);
}
}
Aggregations