| 39 | {} |
| 40 | |
| 41 | void perform_test ( |
| 42 | ) |
| 43 | { |
| 44 | print_spinner(); |
| 45 | |
| 46 | |
| 47 | typedef double scalar_type; |
| 48 | typedef matrix<scalar_type,2,1> sample_type; |
| 49 | |
| 50 | std::vector<sample_type> x; |
| 51 | std::vector<matrix<double,0,1> > x_linearized; |
| 52 | std::vector<scalar_type> y; |
| 53 | |
| 54 | get_checkerboard_problem(x,y, 1000, 2); |
| 55 | |
| 56 | random_subset_selector<sample_type> rx; |
| 57 | random_subset_selector<scalar_type> ry; |
| 58 | rx.set_max_size(x.size()); |
| 59 | ry.set_max_size(x.size()); |
| 60 | |
| 61 | dlog << LINFO << "pos labels: "<< sum(mat(y) == +1); |
| 62 | dlog << LINFO << "neg labels: "<< sum(mat(y) == -1); |
| 63 | |
| 64 | for (unsigned long i = 0; i < x.size(); ++i) |
| 65 | { |
| 66 | rx.add(x[i]); |
| 67 | ry.add(y[i]); |
| 68 | } |
| 69 | |
| 70 | const scalar_type gamma = 2.0; |
| 71 | |
| 72 | typedef radial_basis_kernel<sample_type> kernel_type; |
| 73 | |
| 74 | krr_trainer<kernel_type> krr_trainer; |
| 75 | krr_trainer.use_classification_loss_for_loo_cv(); |
| 76 | krr_trainer.set_kernel(kernel_type(gamma)); |
| 77 | krr_trainer.set_basis(randomly_subsample(x, 100)); |
| 78 | probabilistic_decision_function<kernel_type> df; |
| 79 | |
| 80 | dlog << LINFO << "cross validation: " << cross_validate_trainer(krr_trainer, rx,ry, 4); |
| 81 | print_spinner(); |
| 82 | |
| 83 | running_stats<scalar_type> rs_pos, rs_neg; |
| 84 | |
| 85 | print_spinner(); |
| 86 | df = probabilistic(krr_trainer,3).train(x, y); |
| 87 | for (unsigned long i = 0; i < x.size(); ++i) |
| 88 | { |
| 89 | if (y[i] > 0) |
| 90 | rs_pos.add(df(x[i])); |
| 91 | else |
| 92 | rs_neg.add(df(x[i])); |
| 93 | } |
| 94 | dlog << LINFO << "rs_pos.mean(): "<< rs_pos.mean(); |
| 95 | dlog << LINFO << "rs_neg.mean(): "<< rs_neg.mean(); |
| 96 | DLIB_TEST_MSG(rs_pos.mean() > 0.95, rs_pos.mean()); |
| 97 | DLIB_TEST_MSG(rs_neg.mean() < 0.05, rs_neg.mean()); |
| 98 | rs_pos.clear(); |
nothing calls this directly
no test coverage detected