public static void main(String[] args) throws IOException, InterruptedException {
String category = "";
String training = "";
String testing = "";
for (int i = 0; i < args.length; i++) {
if (args[i].equals("-c")) {
category = args[i + 1];
}
if (args[i].equals("-train")) {
training = args[i + 1];
}
if (args[i].equals("-test")) {
testing = args[i + 1];
}
}
System.out.println("Removing temp files...");
File folder = new File(".");
for (File f : folder.listFiles()) {
if (f.getName().endsWith(".data")) {
f.delete(); // may fail mysteriously - returns boolean you may want to check
}
}
// MaxentTagger tagger = new MaxentTagger("taggers/english-left3words-distsim.tagger");
Documents doc = new Documents();
Instances ins = new Instances();
String[] files = new String[2];
files[0] = training;
files[1] = testing;
Map<String, String[]> alltext = doc.getTextFromFile(files, training, testing);
Map<String, String[]> alltextOriginal = doc.getTextFromFile(files, training, testing);
ArrayList<String> dictionary = ins.dictionary(alltext);
// ArrayList<String> dictionary = ins.dictionaryWithPosTag(alltext,tagger);
String[] categories = new String[1];
categories[0] = category;
int initialTestDataSize = ins.getTestDataSize(alltext); // the inistial size of testing data
Map<String, Integer> finalPredictionLabels =
new HashMap<
String, Integer>(); // A map to store the final predicted labels for evaluation purpose
// Build a model on each defect category
int positive = 0;
int negative = 0;
for (String c : categories) {
// These two lines generate the initial training and testing sets for the current defect
// category
ins.generateTrainingFile(dictionary, alltext, c);
ins.generateTestingFile(dictionary, alltext, c);
// define a counter for iteration number
int iterationNum = 1;
while (ins.getTestDataSize(alltext) != 0) {
System.out.println(
"Iteration " + iterationNum + "...(" + initialTestDataSize + " iterations left)");
svm_iteration(c); // Iterative learning
// Create a map to store the probabilities the SVM classifier output on the test data in
// each iteration
Map<String, Double> predictedProbabilityMap = ins.getPredictedProbability(alltext, c);
// Create a map to store the labels the SVM classifier predicted on the test data in each
// iteration
Map<String, Integer> predictedLabelsMap = ins.getPredictedLabels(alltext, c);
// The selected instance with highest probability
String[] selectedKey_type = ins.selectInstanceHighest(predictedProbabilityMap);
int selectedLabel = 0;
if (selectedKey_type[1].equals("keep")) {
System.out.println("Skip adding");
selectedLabel = predictedLabelsMap.get(selectedKey_type[0]);
alltext =
ins.updateAllTextNoSelection(
alltext, selectedKey_type[0], selectedLabel, dictionary, c);
}
if (selectedKey_type[1].equals("add")) {
selectedLabel = predictedLabelsMap.get(selectedKey_type[0]);
System.out.println("adding instance " + selectedLabel);
alltext = ins.updateAllText(alltext, selectedKey_type[0], selectedLabel, dictionary, c);
if (selectedLabel == 1) positive++;
if (selectedLabel == -1) negative++;
// alltext = ins.updateAllTextNoSelection(alltext, selectedKey_type[0],
// selectedLabel,dictionary, c);
}
/*if(selectedKey_type[1].equals("add")) {
selectedLabel = predictedLabelsMap.get(selectedKey_type[0]);
if(selectedLabel == 1){
System.out.println("adding instance " + selectedLabel);
alltext = ins.updateAllText(alltext, selectedKey_type[0], selectedLabel,dictionary, c);
}
else{
System.out.println("Skip adding");
alltext = ins.updateAllTextNoSelection(alltext, selectedKey_type[0], selectedLabel,dictionary, c);
}
if(selectedLabel == 1) positive++;
if(selectedLabel == -1) negative++;
}*/
// The predicted label of the selected instance above
// int selectedLabel = predictedLabelsMap.get(selectedKey);
// Add the selected instance to the final prediction
finalPredictionLabels.put(selectedKey_type[0], selectedLabel);
/*Update "alltext" map after converting the selected testing instance (with highest probability) to training instance,
whereThe label of selected instance is changed to "train" from "test".
*/
// alltext = ins.updateAllText(alltext, selectedKey, selectedLabel,dictionary, c);
// Generate training and testing files again
ins.generateTrainingFile(dictionary, alltext, c);
ins.generateTestingFile(dictionary, alltext, c);
// Update iteration number and testing data size
iterationNum++;
initialTestDataSize--;
System.out.println(" Label+ : " + positive + " Label- : " + negative);
evaluate(alltextOriginal, finalPredictionLabels, c);
}
}
}
