/**
* Save the dataset to the given stream.
*
* @param os an output stream to where the dataset is saved.
*/
public void save(PrintStream ps) {
for (int idxExample = 0; idxExample < getNumberOfExamples(); ++idxExample) {
PQInput2 input = inputExamples[idxExample];
PQOutput2 output = outputExamples[idxExample];
// The sentence identifier string.
ps.print(input.getId());
int numberOfQuotations = input.getNumberOfQuotations();
for (int quotationIdx = 0; quotationIdx < numberOfQuotations; ++quotationIdx) {
// Quotation separator.
ps.print("§");
int numberOfCoreferences = input.getNumberOfCoreferences(quotationIdx);
for (int coreferenceIdx = 0; coreferenceIdx < numberOfCoreferences; ++coreferenceIdx) {
// Coreference features.
for (int ftr : input.getFeatureCodes(quotationIdx, coreferenceIdx))
ps.print(featureEncoding.getValueByCode(ftr) + " ");
// Coreference separator.
ps.print("\t");
}
// Quotation author.
ps.println(featureEncoding.getValueByCode(output.getAuthor(quotationIdx)));
}
// Next line for the next example.
ps.println();
}
}
/**
* Add the examples in the given dataset to this dataset.
*
* @param other
*/
public void add(PQDataset2 other) throws DatasetException {
if (!featureEncoding.equals(other.featureEncoding))
throw new DatasetException("Different encodings");
// Alloc room to store both datasets (this one and the given one).
PQInput2[] newInputExamples = new PQInput2[inputExamples.length + other.inputExamples.length];
PQOutput2[] newOutputExamples =
new PQOutput2[outputExamples.length + other.outputExamples.length];
// Copy (only reference) the examples in this dataset to the new arrays.
int idx = 0;
for (; idx < inputExamples.length; ++idx) {
newInputExamples[idx] = inputExamples[idx];
newOutputExamples[idx] = outputExamples[idx];
}
// Copy (only reference) the examples in the given dataset to the new
// arrays.
for (int idxO = 0; idxO < other.inputExamples.length; ++idxO, ++idx) {
newInputExamples[idx] = other.inputExamples[idxO];
newOutputExamples[idx] = other.outputExamples[idxO];
}
// Adjust the pointers of this dataset to the new arrays.
this.inputExamples = newInputExamples;
this.outputExamples = newOutputExamples;
}
/**
* Return the number of symbols in the dataset feature-value encoding object. In general, this
* corresponds to the total number of different symbols in the dataset, but can be a different
* number if the encoding was used by other code despite this dataset.
*
* @return
*/
public int getNumberOfSymbols() {
return featureEncoding.size();
}
/**
* Parse the given string and load an example.
*
* @param buff a string that contains an example.
* @return <code>true</code> if the given string is a valid example.
* @throws DatasetException if there is some format problem with the given string.
*/
public boolean parseExample(
Collection<PQInput2> exampleInputs, Collection<PQOutput2> exampleOutputs, String buff)
throws DatasetException {
// Split quotations.
String quotationsInput[] = buff.split("§");
if (quotationsInput.length == 0) return false;
// First field: document ID.
if (quotationsInput[0].trim().length() == 0) return false;
String docId = quotationsInput[0];
LinkedList<LinkedList<LinkedList<Integer>>> exampleInputAsList =
new LinkedList<LinkedList<LinkedList<Integer>>>();
LinkedList<Integer> exampleOutputAsList = new LinkedList<Integer>();
LinkedList<Quotation> quotationsAsList = new LinkedList<Quotation>();
// Walk into the document quotations.
for (int idxQuote = 1; idxQuote < quotationsInput.length; idxQuote++) {
// Split candidate coreferences of the given quotation.
String coreferences[] = quotationsInput[idxQuote].split("\\t");
if (coreferences.length == 0) return false;
// First field: quotation start index, quotation end index,
// right coreference index.
String firstField[] = coreferences[0].split("[ ]");
if (firstField.length != 3) return false;
// Quotation start index, quotation end index.
if ((firstField[0].trim().length() == 0) || (firstField[1].trim().length() == 0))
return false;
Quotation quotation = new Quotation(coreferences.length - 1);
quotation.setQuotationIndex(Integer.parseInt(firstField[0]), Integer.parseInt(firstField[1]));
// Right coreference index.
if (firstField[2].trim().length() == 0) return false;
int rightCoref = Integer.parseInt(firstField[2]) + 1;
// Walk into the coreference feature list.
LinkedList<LinkedList<Integer>> corefFeatureList = new LinkedList<LinkedList<Integer>>();
for (int idxCoref = 1; idxCoref < coreferences.length; ++idxCoref) {
String coreference = coreferences[idxCoref];
// Parse the given coreference features.
String[] features = coreference.split("[ ]");
if (firstField.length < 2) return false;
// First field: coreference start index, coreference end index.
if ((features[0].trim().length() == 0) || (features[1].trim().length() == 0)) return false;
quotation.setCoreferenceIndex(
idxCoref - 1, Integer.parseInt(features[0]), Integer.parseInt(features[1]));
// Encode the features.
LinkedList<Integer> featureList = new LinkedList<Integer>();
for (int idxFtr = 2; idxFtr < features.length; ++idxFtr) {
int code = featureEncoding.put(features[idxFtr]);
if (code >= 0) featureList.add(code);
}
corefFeatureList.add(featureList);
}
// Example input.
exampleInputAsList.add(corefFeatureList);
// Example output.
exampleOutputAsList.add(rightCoref);
// Quotation index information.
quotationsAsList.add(quotation);
}
// Store the loaded example.
if (training) {
/*
* Training examples must store internally their indexes in the
* array of training examples.
*/
exampleInputs.add(new PQInput2(docId, exampleInputAsList, quotationsAsList));
exampleOutputs.add(new PQOutput2(exampleOutputAsList));
} else {
exampleInputs.add(new PQInput2(docId, exampleInputAsList, quotationsAsList));
exampleOutputs.add(new PQOutput2(exampleOutputAsList));
}
return true;
}