void learning() {
int data_sz = samples.size();
int data_dim = samples[0].size();
theta = random_double_list(data_dim);
vector<int> idx;
for(int i = 0; i < data_sz; ++i) {
idx.push_back(i);
}
double coff2 = 2. * beta * alpha;
// training loop
for(int rd = 0; rd < rounds; ++rd) {
random_shuffle(idx.begin(), idx.end());
vector<double> delta(data_dim, 0);
for(auto id : idx) {
double opt1 = alpha * (labels[id] - lr_hypothesis(samples[id]));
for(int i = 0; i < data_dim; ++i) {
double t = opt1 * samples[id][i] - coff2 * theta[i];
delta[i] += t;
}
} // traverse
for(int i = 0; i < data_dim; ++i) {
theta[i] += delta[i];
}
if(debug) {
loss_error.push_back(calc_loss());
}
std::cout << "round " << rd << " finished." << std::endl;
} // training loop
}
void learning() {
std::vector<double> delta_p(k), delta_q(k);
for(auto & kv : usr_dct) {
p[kv.second] = random_double_list(k, 0.1);
usr_bias[kv.second] = 0.1 * random_double();
}
for(auto & kv : item_dct) {
q[kv.second] = random_double_list(k, 0.1);
item_bias[kv.second] = 0.1 * random_double();
}
std::cout << "init done" << std::endl;
auto learning_lambda = [&] (const std::string & uid,
const std::string & iid,
double rating) {
double e = rating - estimate(uid, iid);
// compute delta
for(int i = 0; i < k; ++i) {
delta_p[i] = alpha * (2 * e * q[iid][i] - beta * p[uid][i]);
delta_q[i] = alpha * (2 * e * p[uid][i] - beta * q[iid][i]);
}
// update with delta
for(int i = 0; i < k; ++i) {
p[uid][i] += delta_p[i];
q[iid][i] += delta_q[i];
}
// update bias
usr_bias[uid] += alpha * (2 * e - beta * usr_bias[uid]);
item_bias[iid] += alpha * (2 * e - beta * item_bias[iid]);
};
// learning
for(int rd = 0; rd < rounds; ++rd) {
std::cout << "rd:" << rd << " started" << std::endl;
rating_graph.traverse(learning_lambda);
} // end for
}
