Function dotest_sparse_vectors

dlib/test/ranking.cpp:226–302

// ----------------------------------------------------------------------------------------

    void dotest_sparse_vectors()
    {
        print_spinner();
        dlog << LINFO << "in dotest_sparse_vectors()";

        typedef std::map<unsigned long,double> sample_type;

        typedef sparse_linear_kernel<sample_type> kernel_type;

        svm_rank_trainer<kernel_type> trainer;


        std::vector<ranking_pair<sample_type> > samples;

        ranking_pair<sample_type> p;
        sample_type samp;

        samp[3] = 1; p.relevant.push_back(samp); samp.clear();
        samp[0] = 1; p.nonrelevant.push_back(samp); samp.clear();
        samples.push_back(p);

        samp[2] = 1; p.relevant.push_back(samp); samp.clear();
        samp[0] = 1; p.nonrelevant.push_back(samp); samp.clear();
        samp[1] = 1; p.nonrelevant.push_back(samp); samp.clear();
        samp[1] = 1; p.nonrelevant.push_back(samp); samp.clear();
        samples.push_back(p);


        trainer.set_c(10);

        decision_function<kernel_type> df = trainer.train(samples);

        matrix<double,1,2> res;
        res = 1,1;

        dlog << LINFO << "accuracy: "<< test_ranking_function(df, samples);
        DLIB_TEST(equal(test_ranking_function(df, samples), res));

        DLIB_TEST(equal(test_ranking_function(trainer.train(samples[1]), samples), res));

        trainer.set_epsilon(1e-13);
        df = trainer.train(samples);

        dlog << LINFO << sparse_to_dense(df.basis_vectors(0));
        sample_type truew;
        truew[0] = -0.5;
        truew[1] = -0.5;
        truew[2] =  0.5;
        truew[3] =  0.5;
        DLIB_TEST(length(subtract(truew , df.basis_vectors(0))) < 1e-10);

        dlog << LINFO << "accuracy: "<< test_ranking_function(df, samples);
        DLIB_TEST(equal(test_ranking_function(df, samples), res));

        dlog << LINFO << "cv-accuracy: "<< cross_validate_ranking_trainer(trainer, samples,2);
        DLIB_TEST(std::abs(cross_validate_ranking_trainer(trainer, samples,2)(0) - 0.7777777778) < 0.0001);

        trainer.set_learns_nonnegative_weights(true);