/**
* Load the parameters in the {@code TrainingParameters} file.
*
* @param paramFile the parameter file
* @param supportSequenceTraining wheter sequence training is supported
* @return the parameters
*/
private static TrainingParameters loadTrainingParameters(
final String paramFile, final boolean supportSequenceTraining) {
TrainingParameters params = null;
if (paramFile != null) {
checkInputFile("Training Parameter", new File(paramFile));
InputStream paramsIn = null;
try {
paramsIn = new FileInputStream(new File(paramFile));
params = new opennlp.tools.util.TrainingParameters(paramsIn);
} catch (IOException e) {
throw new TerminateToolException(
-1, "Error during parameters loading: " + e.getMessage(), e);
} finally {
try {
if (paramsIn != null) {
paramsIn.close();
}
} catch (IOException e) {
System.err.println("Error closing the input stream");
}
}
if (!TrainerFactory.isValid(params.getSettings())) {
throw new TerminateToolException(
1, "Training parameters file '" + paramFile + "' is invalid!");
}
}
return params;
}
public static POSModel train(
String languageCode,
ObjectStream<POSSample> samples,
TrainingParameters trainParams,
POSTaggerFactory posFactory)
throws IOException {
String beamSizeString = trainParams.getSettings().get(BeamSearch.BEAM_SIZE_PARAMETER);
int beamSize = POSTaggerME.DEFAULT_BEAM_SIZE;
if (beamSizeString != null) {
beamSize = Integer.parseInt(beamSizeString);
}
POSContextGenerator contextGenerator = posFactory.getPOSContextGenerator();
Map<String, String> manifestInfoEntries = new HashMap<String, String>();
TrainerType trainerType = TrainerFactory.getTrainerType(trainParams.getSettings());
MaxentModel posModel = null;
SequenceClassificationModel<String> seqPosModel = null;
if (TrainerType.EVENT_MODEL_TRAINER.equals(trainerType)) {
ObjectStream<Event> es = new POSSampleEventStream(samples, contextGenerator);
EventTrainer trainer =
TrainerFactory.getEventTrainer(trainParams.getSettings(), manifestInfoEntries);
posModel = trainer.train(es);
} else if (TrainerType.EVENT_MODEL_SEQUENCE_TRAINER.equals(trainerType)) {
POSSampleSequenceStream ss = new POSSampleSequenceStream(samples, contextGenerator);
EventModelSequenceTrainer trainer =
TrainerFactory.getEventModelSequenceTrainer(
trainParams.getSettings(), manifestInfoEntries);
posModel = trainer.train(ss);
} else if (TrainerType.SEQUENCE_TRAINER.equals(trainerType)) {
SequenceTrainer trainer =
TrainerFactory.getSequenceModelTrainer(trainParams.getSettings(), manifestInfoEntries);
// TODO: This will probably cause issue, since the feature generator uses the outcomes array
POSSampleSequenceStream ss = new POSSampleSequenceStream(samples, contextGenerator);
seqPosModel = trainer.train(ss);
} else {
throw new IllegalArgumentException("Trainer type is not supported: " + trainerType);
}
if (posModel != null) {
return new POSModel(languageCode, posModel, beamSize, manifestInfoEntries, posFactory);
} else {
return new POSModel(languageCode, seqPosModel, manifestInfoEntries, posFactory);
}
}
/**
* @deprecated use {@link #train(String, ObjectStream, TrainingParameters, POSTaggerFactory)}
* instead and pass in a {@link POSTaggerFactory} and a {@link TrainingParameters}.
*/
@Deprecated
public static POSModel train(
String languageCode,
ObjectStream<POSSample> samples,
ModelType modelType,
POSDictionary tagDictionary,
Dictionary ngramDictionary,
int cutoff,
int iterations)
throws IOException {
TrainingParameters params = new TrainingParameters();
params.put(TrainingParameters.ALGORITHM_PARAM, modelType.toString());
params.put(TrainingParameters.ITERATIONS_PARAM, Integer.toString(iterations));
params.put(TrainingParameters.CUTOFF_PARAM, Integer.toString(cutoff));
return train(languageCode, samples, params, tagDictionary, ngramDictionary);
}
