/**
* Bigrams encode indices of trees, rather than the String representations themselves, i.e., "3
* 56", denotes the prevElemTree indexed at position 3, and the elemTree at position 56 of their
* corresponding indexers.
*
* @param input
* @param unlex
* @return
*/
private String toStringBigram(
String input, boolean unlex, DiscriminativeFeatureIndexers featureIndexers) {
if (input.equals("0") || input.equals("U")) return "U";
StringBuilder str = new StringBuilder();
String[] inputAr = input.split(" ");
if (inputAr[0].equals("U")) str.append(inputAr[0]);
else {
int prevId = Integer.valueOf(inputAr[0]);
str.append(
unlex
? featureIndexers.getValueOfFeature(ExtractFeatures.FEAT_PREV_ELEM_TREE_UNLEX, prevId)
: featureIndexers.getValueOfFeature(ExtractFeatures.FEAT_PREV_ELEM_TREE, prevId));
}
str.append(" ");
if (inputAr[1].equals("U")) {
str.append(inputAr[1]);
} else {
int curId = Integer.valueOf(inputAr[1]);
str.append(
unlex
? featureIndexers.getValueOfFeature(ExtractFeatures.FEAT_ELEM_TREE_UNLEX, curId)
: featureIndexers.getValueOfFeature(ExtractFeatures.FEAT_ELEM_TREE, curId));
}
return str.toString();
}
/**
* Input srl triple is encoded with ids, to save space, in the format role_id, arg_id, pred_id.
*
* @param input
* @param unlex
* @return
*/
private String toStringSrlTriple(
String input, boolean unlex, DiscriminativeFeatureIndexers featureIndexers) {
if (input.equals("0")) return "U";
String[] ar = input.split(",");
if (!unlex)
return String.format(
"<%s,%s,%s>",
featureIndexers.getRole(Integer.valueOf(ar[0])),
featureIndexers.getWord(Integer.valueOf(ar[1])),
featureIndexers.getWord(Integer.valueOf(ar[2])));
else
return String.format(
"<%s,%s,%s>",
featureIndexers.getRole(Integer.valueOf(ar[0])),
featureIndexers.getPos(Integer.valueOf(ar[1])),
featureIndexers.getPos(Integer.valueOf(ar[2])));
}
public void setNeighboursL2(
String[] neighboursStr, DiscriminativeFeatureIndexers featureIndexers, boolean train) {
this.neighboursL2 = new int[neighboursStr.length];
for (int i = 0; i < neighboursStr.length; i++)
neighboursL2[i] =
featureIndexers.getIndexOfFeature(
ExtractFeatures.FEAT_NEIGHBOURS_L2, neighboursStr[i], train);
}
public void setSrlTriplesPos(
String[] srlTriplesPosStr, DiscriminativeFeatureIndexers featureIndexers, boolean train) {
this.srlTriplesPos = new int[srlTriples.length];
for (int i = 0; i < srlTriplesPosStr.length; i++)
srlTriplesPos[i] =
featureIndexers.getIndexOfFeature(
ExtractFeatures.FEAT_SRL_TRIPLES_POS, srlTriplesPosStr[i], train);
}
public String[] srlTriplesFull(DiscriminativeFeatureIndexers featureIndexers) {
String[] out = new String[srlTriples.length];
for (int i = 0; i < srlTriples.length; i++) {
String srlTriple =
featureIndexers.getValueOfFeature(ExtractFeatures.FEAT_SRL_TRIPLES, srlTriples[i]);
if (srlTriple.equals("0")) {
out[i] = "U";
} else {
String[] srlTripleAr = srlTriple.split(",");
String[] srlTriplePosAr =
featureIndexers
.getValueOfFeature(ExtractFeatures.FEAT_SRL_TRIPLES_POS, srlTriplesPos[i])
.split(",");
out[i] =
String.format(
"<%s,%s/%s,%s/%s>:%s",
featureIndexers.getRole(Integer.valueOf(srlTripleAr[0])), // role
featureIndexers.getWord(Integer.valueOf(srlTripleAr[1])),
featureIndexers.getPos(Integer.valueOf(srlTriplePosAr[1])), // arg
featureIndexers.getWord(Integer.valueOf(srlTripleAr[2])),
featureIndexers.getPos(Integer.valueOf(srlTriplePosAr[2])), // pred
argComplete.get(i));
}
}
return out;
}
private String toStringFeatureWeight(
int vecId,
int featureId,
DiscriminativeFeatureIndexers featureIndexers,
DiscriminativeParams params) {
return String.format(
"%s\t:\t%s",
featureIndexers.getValueOfFeature(vecId, featureId),
params.getParamVecWeight(vecId, featureId));
}
public void setElemTreeUnlex(
String elemTreeUnlexStr, DiscriminativeFeatureIndexers featureIndexers, boolean train) {
this.elemTreeUnlex =
featureIndexers.getIndexOfFeature(
ExtractFeatures.FEAT_ELEM_TREE_UNLEX, elemTreeUnlexStr, train);
}
public void setElemTreeBigram(
String elemTreeBigramStr, DiscriminativeFeatureIndexers featureIndexers, boolean train) {
this.elemTreeBigram =
featureIndexers.getIndexOfFeature(
ExtractFeatures.FEAT_ELEM_TREE_BIGRAM, elemTreeBigramStr, train);
}
public void setPrevElemTree(
String prevElemTreeStr, DiscriminativeFeatureIndexers featureIndexers, boolean train) {
this.prevElemTree =
featureIndexers.getIndexOfFeature(
ExtractFeatures.FEAT_PREV_ELEM_TREE, prevElemTreeStr, train);
}
public String toString(
DiscriminativeFeatureIndexers featureIndexers, DiscriminativeParams params) {
StringBuilder str = new StringBuilder();
str.append(baselineScore)
.append("\t:\t")
.append(params.getParamVecWeight(ExtractFeatures.FEAT_BASELINE_SCORE, 0))
.append("\n")
.append(baselineWordScore)
.append("\t:\t")
.append(params.getParamVecWeight(ExtractFeatures.FEAT_BASELINE_WORD_SCORE, 0))
.append("\n")
.append(
toStringFeatureWeight(
ExtractFeatures.FEAT_ELEM_TREE, elemTree, featureIndexers, params))
.append("\n")
.append(
toStringFeatureWeight(
ExtractFeatures.FEAT_PREV_ELEM_TREE, prevElemTree, featureIndexers, params))
.append("\n")
.append(
toStringBigram(
featureIndexers.getValueOfFeature(
ExtractFeatures.FEAT_ELEM_TREE_BIGRAM, elemTreeBigram),
false,
featureIndexers))
.append("\t:\t")
.append(params.getParamVecWeight(ExtractFeatures.FEAT_ELEM_TREE_BIGRAM, elemTreeBigram))
.append("\n")
.append(
toStringFeatureWeight(
ExtractFeatures.FEAT_ELEM_TREE_UNLEX, elemTreeUnlex, featureIndexers, params))
.append("\n")
.append(
toStringFeatureWeight(
ExtractFeatures.FEAT_PREV_ELEM_TREE_UNLEX,
prevElemTreeUnlex,
featureIndexers,
params))
.append("\n")
.append(
toStringBigram(
featureIndexers.getValueOfFeature(
ExtractFeatures.FEAT_ELEM_TREE_UNLEX_BIGRAM, elemTreeUnlexBigram),
true,
featureIndexers))
.append("\t:\t")
.append(
params.getParamVecWeight(
ExtractFeatures.FEAT_ELEM_TREE_UNLEX_BIGRAM, elemTreeUnlexBigram))
.append("\n")
.append(
toStringFeatureWeight(
ExtractFeatures.FEAT_INTEGRATION_POINT, integrationPoint, featureIndexers, params))
.append("\n")
.append(
toStringIpTree(
featureIndexers.getValueOfFeature(ExtractFeatures.FEAT_IP_ELEM_TREE, ipElemTree),
false,
featureIndexers))
.append("\n")
.append("\t:\t")
.append(params.getParamVecWeight(ExtractFeatures.FEAT_IP_ELEM_TREE, ipElemTree))
.append("\n")
.append(
toStringIpTree(
featureIndexers.getValueOfFeature(ExtractFeatures.FEAT_IP_ELEM_TREE, ipElemTree),
true,
featureIndexers))
.append("\n")
.append("\t:\t")
.append(params.getParamVecWeight(ExtractFeatures.FEAT_IP_ELEM_TREE_UNLEX, ipElemTreeUnlex))
.append("\n");
for (Integer srlTriple : srlTriples)
str.append(
toStringSrlTriple(
featureIndexers.getValueOfFeature(ExtractFeatures.FEAT_SRL_TRIPLES, srlTriple),
false,
featureIndexers))
.append("\t:\t")
.append(params.getParamVecWeight(ExtractFeatures.FEAT_SRL_TRIPLES, srlTriple))
.append(", ");
str.append("\n");
for (Integer srlTriplePos : srlTriplesPos)
str.append(
toStringSrlTriple(
featureIndexers.getValueOfFeature(
ExtractFeatures.FEAT_SRL_TRIPLES_POS, srlTriplePos),
true,
featureIndexers))
.append("\t:\t")
.append(params.getParamVecWeight(ExtractFeatures.FEAT_SRL_TRIPLES_POS, srlTriplePos))
.append(", ");
// str.append("\n").append(semanticScore);
return str.toString();
}
public void setWordL3(
String wordL3Str, DiscriminativeFeatureIndexers featureIndexers, boolean train) {
this.wordL3 = featureIndexers.getIndexOfFeature(ExtractFeatures.FEAT_WORD_L3, wordL3Str, train);
}
public void setIpElemTree(
String ipElemTreeStr, DiscriminativeFeatureIndexers featureIndexers, boolean train) {
this.ipElemTree =
featureIndexers.getIndexOfFeature(ExtractFeatures.FEAT_IP_ELEM_TREE, ipElemTreeStr, train);
}
public void setHeavy(
String[] heavyStr, DiscriminativeFeatureIndexers featureIndexers, boolean train) {
this.heavy = new int[heavyStr.length];
for (int i = 0; i < heavyStr.length; i++)
heavy[i] = featureIndexers.getIndexOfFeature(ExtractFeatures.FEAT_HEAVY, heavyStr[i], train);
}
public void setCoLenPar(
String coLenParStr, DiscriminativeFeatureIndexers featureIndexers, boolean train) {
this.coLenPar =
featureIndexers.getIndexOfFeature(ExtractFeatures.FEAT_CO_LEN_PAR, coLenParStr, train);
}
public void setIntegrationPoint(
String integrationPointStr, DiscriminativeFeatureIndexers featureIndexers, boolean train) {
this.integrationPoint =
featureIndexers.getIndexOfFeature(
ExtractFeatures.FEAT_INTEGRATION_POINT, integrationPointStr, train);
}