int main(int argc, char* argv[]) {
int ch;
std::string test_file;
std::string model_file;
std::string output_file;
while ((ch = getopt(argc, argv, "t:m:o:h")) != -1) {
switch (ch) {
case 't':
test_file = optarg;
break;
case 'm':
model_file = optarg;
break;
case 'o':
output_file = optarg;
break;
case 'h':
default:
print_usage(argc, argv);
exit(0);
}
}
if (test_file.size() == 0 || model_file.size() == 0 || output_file.size() == 0) {
print_usage(argc, argv);
exit(1);
}
LRModel<double> model;
model.Initialize(model_file.c_str());
double y = 0.;
std::vector<std::pair<size_t, double> > x;
FILE* wfp = fopen(output_file.c_str(), "w");
size_t cnt = 0, correct = 0;
double loss = 0.;
FileParser<double> parser;
parser.OpenFile(test_file.c_str());
std::vector<std::pair<double, unsigned> > pred_scores;
while (1) {
bool res = parser.ReadSample(y, x);
if (!res) break;
double pred = model.Predict(x);
pred = std::max(std::min(pred, 1. - 10e-15), 10e-15);
fprintf(wfp, "%lf\n", pred);
pred_scores.push_back(std::move(
std::make_pair(pred, static_cast<unsigned>(y))));
++cnt;
double pred_label = 0;
if (pred > 0.5) pred_label = 1;
if (util_equal(pred_label, y)) ++correct;
pred = std::max(std::min(pred, 1. - 10e-15), 10e-15);
loss += y > 0 ? -log(pred) : -log(1. - pred);
}
std::sort(
pred_scores.begin(),
pred_scores.end(),
[] (const std::pair<double, unsigned>& l, const std::pair<double, unsigned>& r) {
return l.first > r.first;
}
);
double auc = calc_auc(pred_scores);
if (cnt > 0) {
printf("Accuracy = %.2lf%% (%zu/%zu)\n",
static_cast<double>(correct) / cnt * 100, correct, cnt);
printf("Log-likelihood = %lf\n", loss / cnt);
printf("AUC = %lf\n", auc);
}
parser.CloseFile();
fclose(wfp);
return 0;
}
