1 | // Copyright 2015 Georg-August-Universität Göttingen, Germany
|
---|
2 | //
|
---|
3 | // Licensed under the Apache License, Version 2.0 (the "License");
|
---|
4 | // you may not use this file except in compliance with the License.
|
---|
5 | // You may obtain a copy of the License at
|
---|
6 | //
|
---|
7 | // http://www.apache.org/licenses/LICENSE-2.0
|
---|
8 | //
|
---|
9 | // Unless required by applicable law or agreed to in writing, software
|
---|
10 | // distributed under the License is distributed on an "AS IS" BASIS,
|
---|
11 | // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
---|
12 | // See the License for the specific language governing permissions and
|
---|
13 | // limitations under the License.
|
---|
14 |
|
---|
15 | package de.ugoe.cs.cpdp.eval;
|
---|
16 |
|
---|
17 | import java.io.FileNotFoundException;
|
---|
18 | import java.io.FileOutputStream;
|
---|
19 | import java.io.PrintWriter;
|
---|
20 | import java.util.ArrayList;
|
---|
21 | import java.util.LinkedList;
|
---|
22 | import java.util.List;
|
---|
23 |
|
---|
24 | import de.ugoe.cs.cpdp.training.ITrainer;
|
---|
25 | import de.ugoe.cs.cpdp.training.IWekaCompatibleTrainer;
|
---|
26 | import de.ugoe.cs.util.StringTools;
|
---|
27 | import weka.classifiers.Classifier;
|
---|
28 | import weka.classifiers.Evaluation;
|
---|
29 | import weka.core.Attribute;
|
---|
30 | import weka.core.Instances;
|
---|
31 |
|
---|
32 | /**
|
---|
33 | * Base class for the evaluation of results of classifiers compatible with the {@link Classifier}
|
---|
34 | * interface. For each classifier, the following metrics are calculated:
|
---|
35 | * <ul>
|
---|
36 | * <li>succHe: Success with recall>0.7, precision>0.5</li>
|
---|
37 | * <li>succZi: Success with recall>0.7, precision>0.7</li>
|
---|
38 | * <li>succG75: Success with gscore>0.75</li>
|
---|
39 | * <li>succG60: Success with gscore>0.6</li>
|
---|
40 | * <li>error</li>
|
---|
41 | * <li>recall</li>
|
---|
42 | * <li>precision</li>
|
---|
43 | * <li>fscore</li>
|
---|
44 | * <li>gscore</li>
|
---|
45 | * <li>AUC</li>
|
---|
46 | * <li>AUCEC (weighted by LOC, if applicable; 0.0 if LOC not available)</li>
|
---|
47 | * <li>tpr: true positive rate</li>
|
---|
48 | * <li>tnr: true negative rate</li>
|
---|
49 | * <li>tp: true positives</li>
|
---|
50 | * <li>fp: false positives</li>
|
---|
51 | * <li>tn: true negatives</li>
|
---|
52 | * <li>fn: false negatives</li>
|
---|
53 | * <li>errortrain: training error</li>
|
---|
54 | * <li>recalltrain: training recall</li>
|
---|
55 | * <li>precisiontrain: training precision</li>
|
---|
56 | * <li>succHetrain: training success with recall>0.7 and precision>0.5
|
---|
57 | * </ul>
|
---|
58 | *
|
---|
59 | * @author Steffen Herbold
|
---|
60 | */
|
---|
61 | public abstract class AbstractWekaEvaluation implements IEvaluationStrategy {
|
---|
62 |
|
---|
63 | /**
|
---|
64 | * writer for the evaluation results
|
---|
65 | */
|
---|
66 | private PrintWriter output = new PrintWriter(System.out);
|
---|
67 |
|
---|
68 | private boolean outputIsSystemOut = true;
|
---|
69 |
|
---|
70 | /**
|
---|
71 | * Creates the weka evaluator. Allows the creation of the evaluator in different ways, e.g., for
|
---|
72 | * cross-validation or evaluation on the test data.
|
---|
73 | *
|
---|
74 | * @param testdata
|
---|
75 | * test data
|
---|
76 | * @param classifier
|
---|
77 | * classifier used
|
---|
78 | * @return evaluator
|
---|
79 | */
|
---|
80 | protected abstract Evaluation createEvaluator(Instances testdata, Classifier classifier);
|
---|
81 |
|
---|
82 | /*
|
---|
83 | * (non-Javadoc)
|
---|
84 | *
|
---|
85 | * @see de.ugoe.cs.cpdp.eval.EvaluationStrategy#apply(weka.core.Instances, weka.core.Instances,
|
---|
86 | * java.util.List, boolean)
|
---|
87 | */
|
---|
88 | @Override
|
---|
89 | public void apply(Instances testdata,
|
---|
90 | Instances traindata,
|
---|
91 | List<ITrainer> trainers,
|
---|
92 | boolean writeHeader)
|
---|
93 | {
|
---|
94 | final List<Classifier> classifiers = new LinkedList<Classifier>();
|
---|
95 | for (ITrainer trainer : trainers) {
|
---|
96 | if (trainer instanceof IWekaCompatibleTrainer) {
|
---|
97 | classifiers.add(((IWekaCompatibleTrainer) trainer).getClassifier());
|
---|
98 | }
|
---|
99 | else {
|
---|
100 | throw new RuntimeException("The selected evaluator only support Weka classifiers");
|
---|
101 | }
|
---|
102 | }
|
---|
103 |
|
---|
104 | if (writeHeader) {
|
---|
105 | output.append("version,size_test,size_training");
|
---|
106 | for (ITrainer trainer : trainers) {
|
---|
107 | output.append(",succHe_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
108 | output.append(",succZi_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
109 | output.append(",succG75_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
110 | output.append(",succG60_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
111 | output.append(",error_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
112 | output.append(",recall_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
113 | output.append(",precision_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
114 | output.append(",fscore_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
115 | output.append(",gscore_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
116 | output.append(",mcc_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
117 | output.append(",auc_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
118 | output.append(",aucec_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
119 | output.append(",tpr_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
120 | output.append(",tnr_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
121 | output.append(",tp_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
122 | output.append(",fn_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
123 | output.append(",tn_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
124 | output.append(",fp_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
125 | output.append(",trainerror_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
126 | output.append(",trainrecall_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
127 | output.append(",trainprecision_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
128 | output.append(",trainsuccHe_" + ((IWekaCompatibleTrainer) trainer).getName());
|
---|
129 | }
|
---|
130 | output.append(StringTools.ENDLINE);
|
---|
131 | }
|
---|
132 |
|
---|
133 | output.append(testdata.relationName());
|
---|
134 | output.append("," + testdata.numInstances());
|
---|
135 | output.append("," + traindata.numInstances());
|
---|
136 |
|
---|
137 | Evaluation eval = null;
|
---|
138 | Evaluation evalTrain = null;
|
---|
139 | for (Classifier classifier : classifiers) {
|
---|
140 | eval = createEvaluator(testdata, classifier);
|
---|
141 | evalTrain = createEvaluator(traindata, classifier);
|
---|
142 |
|
---|
143 | double pf =
|
---|
144 | eval.numFalsePositives(1) / (eval.numFalsePositives(1) + eval.numTrueNegatives(1));
|
---|
145 | double gmeasure = 2 * eval.recall(1) * (1.0 - pf) / (eval.recall(1) + (1.0 - pf));
|
---|
146 | double mcc =
|
---|
147 | (eval.numTruePositives(1) * eval.numTrueNegatives(1) - eval.numFalsePositives(1) *
|
---|
148 | eval.numFalseNegatives(1)) /
|
---|
149 | Math.sqrt((eval.numTruePositives(1) + eval.numFalsePositives(1)) *
|
---|
150 | (eval.numTruePositives(1) + eval.numFalseNegatives(1)) *
|
---|
151 | (eval.numTrueNegatives(1) + eval.numFalsePositives(1)) *
|
---|
152 | (eval.numTrueNegatives(1) + eval.numFalseNegatives(1)));
|
---|
153 | double aucec = calculateReviewEffort(testdata, classifier);
|
---|
154 |
|
---|
155 | if (eval.recall(1) >= 0.7 && eval.precision(1) >= 0.5) {
|
---|
156 | output.append(",1");
|
---|
157 | }
|
---|
158 | else {
|
---|
159 | output.append(",0");
|
---|
160 | }
|
---|
161 |
|
---|
162 | if (eval.recall(1) >= 0.7 && eval.precision(1) >= 0.7) {
|
---|
163 | output.append(",1");
|
---|
164 | }
|
---|
165 | else {
|
---|
166 | output.append(",0");
|
---|
167 | }
|
---|
168 |
|
---|
169 | if (gmeasure > 0.75) {
|
---|
170 | output.append(",1");
|
---|
171 | }
|
---|
172 | else {
|
---|
173 | output.append(",0");
|
---|
174 | }
|
---|
175 |
|
---|
176 | if (gmeasure > 0.6) {
|
---|
177 | output.append(",1");
|
---|
178 | }
|
---|
179 | else {
|
---|
180 | output.append(",0");
|
---|
181 | }
|
---|
182 |
|
---|
183 | output.append("," + eval.errorRate());
|
---|
184 | output.append("," + eval.recall(1));
|
---|
185 | output.append("," + eval.precision(1));
|
---|
186 | output.append("," + eval.fMeasure(1));
|
---|
187 | output.append("," + gmeasure);
|
---|
188 | output.append("," + mcc);
|
---|
189 | output.append("," + eval.areaUnderROC(1));
|
---|
190 | output.append("," + aucec);
|
---|
191 | output.append("," + eval.truePositiveRate(1));
|
---|
192 | output.append("," + eval.trueNegativeRate(1));
|
---|
193 | output.append("," + eval.numTruePositives(1));
|
---|
194 | output.append("," + eval.numFalseNegatives(1));
|
---|
195 | output.append("," + eval.numTrueNegatives(1));
|
---|
196 | output.append("," + eval.numFalsePositives(1));
|
---|
197 | output.append("," + evalTrain.errorRate());
|
---|
198 | output.append("," + evalTrain.recall(1));
|
---|
199 | output.append("," + evalTrain.precision(1));
|
---|
200 | if (evalTrain.recall(1) >= 0.7 && evalTrain.precision(1) >= 0.5) {
|
---|
201 | output.append(",1");
|
---|
202 | }
|
---|
203 | else {
|
---|
204 | output.append(",0");
|
---|
205 | }
|
---|
206 | }
|
---|
207 |
|
---|
208 | output.append(StringTools.ENDLINE);
|
---|
209 | output.flush();
|
---|
210 | }
|
---|
211 |
|
---|
212 | private double calculateReviewEffort(Instances testdata, Classifier classifier) {
|
---|
213 |
|
---|
214 | final Attribute loc = testdata.attribute("loc");
|
---|
215 | if (loc == null) {
|
---|
216 | return 0.0;
|
---|
217 | }
|
---|
218 |
|
---|
219 | final List<Integer> bugPredicted = new ArrayList<>();
|
---|
220 | final List<Integer> nobugPredicted = new ArrayList<>();
|
---|
221 | double totalLoc = 0.0d;
|
---|
222 | int totalBugs = 0;
|
---|
223 | for (int i = 0; i < testdata.numInstances(); i++) {
|
---|
224 | try {
|
---|
225 | if (Double.compare(classifier.classifyInstance(testdata.instance(i)), 0.0d) == 0) {
|
---|
226 | nobugPredicted.add(i);
|
---|
227 | }
|
---|
228 | else {
|
---|
229 | bugPredicted.add(i);
|
---|
230 | }
|
---|
231 | }
|
---|
232 | catch (Exception e) {
|
---|
233 | throw new RuntimeException(
|
---|
234 | "unexpected error during the evaluation of the review effort",
|
---|
235 | e);
|
---|
236 | }
|
---|
237 | if (Double.compare(testdata.instance(i).classValue(), 1.0d) == 0) {
|
---|
238 | totalBugs++;
|
---|
239 | }
|
---|
240 | totalLoc += testdata.instance(i).value(loc);
|
---|
241 | }
|
---|
242 |
|
---|
243 | final List<Double> reviewLoc = new ArrayList<>(testdata.numInstances());
|
---|
244 | final List<Double> bugsFound = new ArrayList<>(testdata.numInstances());
|
---|
245 |
|
---|
246 | double currentBugsFound = 0;
|
---|
247 |
|
---|
248 | while (!bugPredicted.isEmpty()) {
|
---|
249 | double minLoc = Double.MAX_VALUE;
|
---|
250 | int minIndex = -1;
|
---|
251 | for (int i = 0; i < bugPredicted.size(); i++) {
|
---|
252 | double currentLoc = testdata.instance(bugPredicted.get(i)).value(loc);
|
---|
253 | if (currentLoc < minLoc) {
|
---|
254 | minIndex = i;
|
---|
255 | minLoc = currentLoc;
|
---|
256 | }
|
---|
257 | }
|
---|
258 | if (minIndex != -1) {
|
---|
259 | reviewLoc.add(minLoc / totalLoc);
|
---|
260 |
|
---|
261 | currentBugsFound += testdata.instance(bugPredicted.get(minIndex)).classValue();
|
---|
262 | bugsFound.add(currentBugsFound);
|
---|
263 |
|
---|
264 | bugPredicted.remove(minIndex);
|
---|
265 | }
|
---|
266 | else {
|
---|
267 | throw new RuntimeException("Shouldn't happen!");
|
---|
268 | }
|
---|
269 | }
|
---|
270 |
|
---|
271 | while (!nobugPredicted.isEmpty()) {
|
---|
272 | double minLoc = Double.MAX_VALUE;
|
---|
273 | int minIndex = -1;
|
---|
274 | for (int i = 0; i < nobugPredicted.size(); i++) {
|
---|
275 | double currentLoc = testdata.instance(nobugPredicted.get(i)).value(loc);
|
---|
276 | if (currentLoc < minLoc) {
|
---|
277 | minIndex = i;
|
---|
278 | minLoc = currentLoc;
|
---|
279 | }
|
---|
280 | }
|
---|
281 | if (minIndex != -1) {
|
---|
282 | reviewLoc.add(minLoc / totalLoc);
|
---|
283 |
|
---|
284 | currentBugsFound += testdata.instance(nobugPredicted.get(minIndex)).classValue();
|
---|
285 | bugsFound.add(currentBugsFound);
|
---|
286 | nobugPredicted.remove(minIndex);
|
---|
287 | }
|
---|
288 | else {
|
---|
289 | throw new RuntimeException("Shouldn't happen!");
|
---|
290 | }
|
---|
291 | }
|
---|
292 |
|
---|
293 | double auc = 0.0;
|
---|
294 | for (int i = 0; i < bugsFound.size(); i++) {
|
---|
295 | auc += reviewLoc.get(i) * bugsFound.get(i) / totalBugs;
|
---|
296 | }
|
---|
297 |
|
---|
298 | return auc;
|
---|
299 | }
|
---|
300 |
|
---|
301 | /*
|
---|
302 | * (non-Javadoc)
|
---|
303 | *
|
---|
304 | * @see de.ugoe.cs.cpdp.Parameterizable#setParameter(java.lang.String)
|
---|
305 | */
|
---|
306 | @Override
|
---|
307 | public void setParameter(String parameters) {
|
---|
308 | if (output != null && !outputIsSystemOut) {
|
---|
309 | output.close();
|
---|
310 | }
|
---|
311 | if ("system.out".equals(parameters) || "".equals(parameters)) {
|
---|
312 | output = new PrintWriter(System.out);
|
---|
313 | outputIsSystemOut = true;
|
---|
314 | }
|
---|
315 | else {
|
---|
316 | try {
|
---|
317 | output = new PrintWriter(new FileOutputStream(parameters));
|
---|
318 | outputIsSystemOut = false;
|
---|
319 | }
|
---|
320 | catch (FileNotFoundException e) {
|
---|
321 | throw new RuntimeException(e);
|
---|
322 | }
|
---|
323 | }
|
---|
324 | }
|
---|
325 | }
|
---|