public OutputData learn(InputData id, Target tg, int k, String task) {
this.k = k;
OutputData ret;
this.col = id.getCol();
this.row = id.getRow();
this.groupNum = id.getGroup();
this.task = task;
this.groupRangeUpperLimit = id.getGroupRangeUpperLimit();
init();
int t = 0;
for (int iter = 0; iter < 100; iter++) { // tmp
System.out.println(iter); // ***********************
double diff = 0;
for (int p = 0; p < id.getRow(); p++, t++) {
// double tmpEta = eater / (t0 + t*0.1);
// double tmpEta = eater / Math.pow((t + 1), power_t);
double eta = 0.005;
Map<Integer, Double> record = id.getOneRecord(p); // pick up one record
double y = tg.getOneTarget(p); // pickup the target for the chosen record
w0 = w0 - eta * (calcGrad(0, record, y, "w0") + 2 * lambda0 * w0);
// System.out.println("w0:" + w0);//**************
for (int i : record.keySet()) {
double gradWi = calcGrad(i, record, y, "w"); // dloss(predict("w", i), y);
int pi = pi(i);
double nextWi = w[i] - eta * (gradWi + 2 * lambdaW.get(pi) * w[i]);
w[i] = nextWi;
for (int f = 0; f < k; f++) {
double gradVij =
dloss(
predict(y, record),
y,
i,
f,
"v",
record); // calcGradV(i, f, pi, y, "v"); // dloss(predict("v", i, f), y)
// System.out.println("key:" + key + ", f:" + f + ", groupOfKey:" + groupOfKey);
// //******************
V[i][f] -= eta * (gradVij + 2 * lambdaV[f][pi] * V[i][f]);
// System.out.println("V[key][f]:" + V[key][f]); //***************
}
}
if (task.equals("regression")) {
double pred = predict(y, record);
double d = y - pred;
diff += d * d;
} else if (task.equals("classification")) {
if (y != predict(y, record)) {
// System.out.println("tg:" + this.tg + ", predict:" + predict()); //****
diff++;
}
}
}
// effect evaluation
results.add(diff);
}
System.out.println(results); // ****
ret = new OutputData(w0, w, V);
return ret;
}
