public static void trainLibLinear() throws Exception {
System.out.println("train SVMs.");
Indexer<String> featureIndexer = IOUtils.readIndexer(WORD_INDEXER_FILE);
List<SparseVector> trainData = SparseVector.readList(TRAIN_DATA_FILE);
List<SparseVector> testData = SparseVector.readList(TEST_DATA_FILE);
Collections.shuffle(trainData);
Collections.shuffle(testData);
// List[] lists = new List[] { trainData, testData };
//
// for (int i = 0; i < lists.length; i++) {
// List<SparseVector> list = lists[i];
// for (int j = 0; j < list.size(); j++) {
// SparseVector sv = list.get(j);
// if (sv.label() > 0) {
// sv.setLabel(1);
// }
// }
// }
Problem prob = new Problem();
prob.l = trainData.size();
prob.n = featureIndexer.size() + 1;
prob.y = new double[prob.l];
prob.x = new Feature[prob.l][];
prob.bias = -1;
if (prob.bias >= 0) {
prob.n++;
}
for (int i = 0; i < trainData.size(); i++) {
SparseVector x = trainData.get(i);
Feature[] input = new Feature[prob.bias > 0 ? x.size() + 1 : x.size()];
for (int j = 0; j < x.size(); j++) {
int index = x.indexAtLoc(j) + 1;
double value = x.valueAtLoc(j);
assert index >= 0;
input[j] = new FeatureNode(index + 1, value);
}
if (prob.bias >= 0) {
input[input.length - 1] = new FeatureNode(prob.n, prob.bias);
}
prob.x[i] = input;
prob.y[i] = x.label();
}
Model model = Linear.train(prob, getSVMParamter());
CounterMap<Integer, Integer> cm = new CounterMap<Integer, Integer>();
for (int i = 0; i < testData.size(); i++) {
SparseVector sv = testData.get(i);
Feature[] input = new Feature[sv.size()];
for (int j = 0; j < sv.size(); j++) {
int index = sv.indexAtLoc(j) + 1;
double value = sv.valueAtLoc(j);
input[j] = new FeatureNode(index + 1, value);
}
double[] dec_values = new double[model.getNrClass()];
Linear.predictValues(model, input, dec_values);
int max_id = ArrayMath.argmax(dec_values);
int pred = model.getLabels()[max_id];
int answer = sv.label();
cm.incrementCount(answer, pred, 1);
}
System.out.println(cm);
model.save(new File(MODEL_FILE));
}
