-
Notifications
You must be signed in to change notification settings - Fork 12
Expand file tree
/
Copy pathCorefDriver.java
More file actions
210 lines (201 loc) · 8.71 KB
/
Copy pathCorefDriver.java
File metadata and controls
210 lines (201 loc) · 8.71 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
package coref;
import edu.illinois.cs.cogcomp.core.datastructures.Pair;
import edu.illinois.cs.cogcomp.sl.core.SLModel;
import edu.illinois.cs.cogcomp.sl.core.SLParameters;
import edu.illinois.cs.cogcomp.sl.core.SLProblem;
import edu.illinois.cs.cogcomp.sl.learner.Learner;
import edu.illinois.cs.cogcomp.sl.learner.LearnerFactory;
import edu.illinois.cs.cogcomp.sl.util.Lexiconer;
import edu.illinois.cs.cogcomp.sl.util.WeightVector;
import edu.stanford.nlp.ling.CoreLabel;
import logic.Verbs;
import structure.Node;
import structure.StanfordProblem;
import structure.StanfordSchema;
import utils.Folds;
import utils.Params;
import utils.Tools;
import java.util.*;
public class CorefDriver {
public static void crossVal(List<StanfordProblem> probs, List<List<Integer>> foldIndices)
throws Exception {
double acc1 = 0.0, acc2 = 0.0;
for(int i=0;i<foldIndices.size(); i++) {
List<Integer> train = new ArrayList<>();
List<Integer> test = new ArrayList<>();
for(int j=0; j<foldIndices.size(); ++j) {
if(i==j) test.addAll(foldIndices.get(j));
else train.addAll(foldIndices.get(j));
}
Pair<Double, Double> pair = doTrainTest(probs, train, test, i);
acc1 += pair.getFirst();
acc2 += pair.getSecond();
}
System.out.println("CV : " + (acc1/foldIndices.size()) + " " + (acc2/foldIndices.size()));
}
public static Pair<Double, Double> doTrainTest(
List<StanfordProblem> probs, List<Integer> trainIndices,
List<Integer> testIndices, int id) throws Exception {
List<List<StanfordProblem>> split = Folds.getDataSplitForStanford(
probs, trainIndices, testIndices, 0.0);
List<StanfordProblem> trainProbs = split.get(0);
List<StanfordProblem> testProbs = split.get(2);
SLProblem train = getSP(trainProbs, true);
SLProblem test = getSP(testProbs, false);
System.out.println("Train : "+train.instanceList.size()+" Test : "+test.instanceList.size());
Params.printLog = false;
trainModel(Params.modelDir+Params.corefPrefix+id+Params.modelSuffix, train);
Params.printLog = true;
return testModel(Params.modelDir+Params.corefPrefix+id+Params.modelSuffix, test);
}
public static SLProblem getSP(List<StanfordProblem> problemList, boolean train)
throws Exception{
SLProblem problem = new SLProblem();
for(StanfordProblem prob : problemList){
if(prob.id == 793 || prob.id == 838 || prob.id == 777 ||
prob.id == 778 || prob.id == 837 || prob.id == 1600 ||
prob.id == 1610 || prob.id == 1232857115 || prob.id == 1232856836 ||
prob.id == 1232856820 || prob.id == 1232856830 || prob.id == 1232857096 ||
prob.id == 1232857165 || prob.id == 1232857252) continue;
logic.LogicX x = new logic.LogicX(prob);
logic.LogicY y = new logic.LogicY(x, prob.expr, prob.rates);
List<Node> nodes = y.expr.getAllSubNodes();
List<CorefX> xList = new ArrayList<>();
List<CorefY> yList = new ArrayList<>();
boolean reasonsFound = true;
for(Node node : nodes) {
if(node.children.size() == 0) continue;
int quantLeft = node.children.get(0).quantIndex < node.children.get(1).quantIndex ?
node.children.get(0).quantIndex : node.children.get(1).quantIndex;
int quantRight = node.children.get(0).quantIndex < node.children.get(1).quantIndex ?
node.children.get(1).quantIndex : node.children.get(0).quantIndex;
CorefX logicX = new CorefX(prob, quantLeft, quantRight, node.infRuleType);
String label = node.label;
if(node.children.get(0).quantIndex >= node.children.get(1).quantIndex &&
node.label.equals("DIV")) {
label += "_REV";
}
CorefY logicY = new CorefY(label, null);
xList.add(logicX);
yList.add(logicY);
reasonsFound = true;
if(node.infRuleType == null) {
reasonsFound = false;
System.out.println("==========================================");
System.out.println(prob.id+" : "+prob.question);
System.out.println();
for(StanfordSchema schema : prob.schema) {
System.out.println(schema);
System.out.println("VerbCat:"+ Tools.getKeyForMaxValue(Verbs.verbClassify(
prob.tokens.get(schema.sentId).get(schema.verb).lemma(),
Tools.spanToLemmaList(prob.tokens.get(schema.sentId), schema.unit))));
}
System.out.println(prob.questionSchema);
System.out.println();
System.out.println("Quantities : "+prob.quantities);
System.out.println("Quant of Interest: "+quantLeft+" "+quantRight);
System.out.println();
System.out.println("==========================================");
break;
}
}
if(reasonsFound) {
for(int i=0; i<xList.size(); ++i) {
problem.addExample(xList.get(i), yList.get(i));
}
}
}
return problem;
}
public static Pair<Double, Double> testModel(String modelPath, SLProblem sp)
throws Exception {
SLModel model = SLModel.loadModel(modelPath);
Set<Integer> incorrect = new HashSet<>();
Set<Integer> total = new HashSet<>();
double acc = 0.0;
for (int i = 0; i < sp.instanceList.size(); i++) {
System.out.println("==========================================");
CorefX prob = (CorefX) sp.instanceList.get(i);
CorefY gold = (CorefY) sp.goldStructureList.get(i);
CorefY pred = (CorefY) model.infSolver.getBestStructure(model.wv, prob);
total.add(prob.problemId);
boolean correct = false;
if(CorefY.getLoss(gold, pred) < 0.0001) {
acc += 1;
correct = true;
} else {
incorrect.add(prob.problemId);
}
if((correct && Params.printCorrect) ||
(!correct && Params.printMistakes)){
System.out.println(prob.problemId+" : "+prob.text);
for(List<CoreLabel> tokens : prob.tokens) {
for(CoreLabel token : tokens) {
System.out.print(token.lemma()+"/"+token.tag()+" ");
}
}
System.out.println();
for(StanfordSchema schema : prob.schema) {
System.out.println(schema);
System.out.println("VerbCat:"+ Tools.getKeyForMaxValue(Verbs.verbClassify(
prob.tokens.get(schema.sentId).get(schema.verb).lemma(),
Tools.spanToLemmaList(prob.tokens.get(schema.sentId), schema.unit))));
}
System.out.println(prob.questionSchema);
System.out.println("Wordnet: "+Arrays.asList(prob.wordnetRelations));
System.out.println();
System.out.println("InferenceType: "+prob.infType);
System.out.println("Quantities : "+prob.quantities);
System.out.println("Quant of Interest: "+prob.quantIndex1+" "+prob.quantIndex2);
System.out.println();
System.out.println("Gold : "+gold);
System.out.println("Pred : "+pred);
System.out.println("Loss : "+ CorefY.getLoss(gold, pred));
System.out.println();
}
System.out.println("==========================================");
}
System.out.println("Accuracy : = " + acc + " / " + sp.instanceList.size()
+ " = " + (acc/sp.instanceList.size()));
System.out.println("Strict Accuracy : ="+ (1-1.0*incorrect.size()/total.size()));
System.out.println("Incorrect Ids: "+ Arrays.asList(incorrect));
return new Pair<>(acc/sp.instanceList.size(), 1-1.0*incorrect.size()/total.size());
}
public static void trainModel(String modelPath, SLProblem train)
throws Exception {
SLModel model = new SLModel();
Lexiconer lm = new Lexiconer();
lm.setAllowNewFeatures(true);
model.lm = lm;
CorefFeatGen fg = new CorefFeatGen(lm);
model.featureGenerator = fg;
model.infSolver = new CorefInfSolver(fg);
SLParameters para = new SLParameters();
para.loadConfigFile(Params.spConfigFile);
para.MAX_NUM_ITER = 5;
Learner learner = LearnerFactory.getLearner(model.infSolver, fg, para);
model.wv = latentSVMLearner(learner, train, (CorefInfSolver) model.infSolver, 5);
lm.setAllowNewFeatures(false);
model.saveModel(modelPath);
}
public static WeightVector latentSVMLearner(
Learner learner, SLProblem sp, CorefInfSolver infSolver,
int maxIter) throws Exception {
WeightVector wv = new WeightVector(7000);
wv.setExtendable(true);
for(int i=0; i<maxIter; ++i) {
System.err.println("Latent SSVM : Iteration "+i);
SLProblem newProb = new SLProblem();
for(int j=0; j<sp.goldStructureList.size(); ++j) {
CorefX prob = (CorefX) sp.instanceList.get(j);
CorefY gold = (CorefY) sp.goldStructureList.get(j);
CorefY bestLatent = infSolver.getBestStructure(prob, gold, wv, true);
newProb.addExample(prob, bestLatent);
}
System.err.println("Learning SSVM");
wv = learner.train(newProb, wv);
System.err.println("Done");
}
return wv;
}
}