lpboost.cpp

Go to the documentation of this file.

#include <assert.h>
#include <cmath>
#include <iostream>
#include "lpboost.h"
extern "C"{
#include <glpk.h>
}

REGISTER_CREATOR(lemga::LPBoost);

namespace lemga {

#define U(i) ((i)+1)  //U(0) to U(n_samples-1)
#define R(t) ((t)+1)  //R(0) to R(T-1)

REAL LPBoost::train () {
    assert(n_in_agg == 0 && empty());
    assert(ptd != 0 && ptw != 0);
    assert(lm_base != 0); // we need lm_base to create new hypotheses
    assert(!grad_desc_view);

    // Construct inner problem
    LPX* lp = lpx_create_prob();
    lpx_add_cols(lp, n_samples);                        // u_i
    for (UINT i = 0; i < n_samples; ++i) {
        lpx_set_col_bnds(lp, U(i), LPX_DB, 0.0,
                         RegC * (*ptw)[i] * n_samples); // 0 <= u_i <= C_i
        lpx_set_obj_coef(lp, U(i), -1);                 // obj: -sum u_i
    }
    lpx_set_obj_dir(lp, LPX_MIN);                       // min obj

    // For adding columns
    int* ndx = new int[n_samples+1]; double* val = new double[n_samples+1];

    REAL besterr = HUGE_VAL;
    pDataWgt pdw = ptw;

    for (UINT t = 0; t < max_n_model; ++t) {
        const pLearnModel p = train_with_smpwgt(pdw);

        REAL err = 0;
        for (UINT i = 0; i < n_samples; ++i) {
            if (p->c_error(p->get_output(i), ptd->y(i)) > 0.1)
                err += (*pdw)[i];
        }
        if (err >= besterr - EPSILON) // Cannot find better hypotheses
            break;

        // Add one more constraint R(t) = -sum u_i y_i h_t(x_i) >= -1
        lpx_add_rows(lp, 1);
        for (UINT i = 0; i < n_samples; ++i) {
            ndx[i+1] = U(i);
            val[i+1] = - p->get_output(i)[0] * ptd->y(i)[0];
        }
        lpx_set_mat_row(lp, R(t), n_samples, ndx, val);
        lpx_set_row_bnds(lp, R(t), LPX_LO, -1.0, 0.0);  // R(t) >= -1

        // Solve inner problem
        lpx_simplex(lp);
        REAL sumu = -lpx_get_obj_val(lp);
        if (sumu < EPSILON) { // we do not expect this to happen
            std::cerr << "Warning: sum u is " << sumu << "; quit earlier.\n";
            break;
        }
        besterr = (1.0 - 1.0 / sumu) / 2.0;

        lm.push_back(p); lm_wgt.push_back(0);
        ++n_in_agg;

        // Update sample weights
        DataWgt* sample_wgt = new DataWgt(n_samples);
        for (UINT i = 0; i < n_samples; ++i) {
            double wgt;
            wgt = lpx_get_col_prim(lp, U(i));
            assert(wgt >= -EPSILON);
            if (wgt < 0) wgt = 0;
            (*sample_wgt)[i] = wgt / sumu;
        }
        pdw = sample_wgt;

        // Update hypothesis coefficients
        for (UINT k = 0; k <= t; ++k) {
            lm_wgt[k] = lpx_get_row_dual(lp, R(k));
            assert(lm_wgt[k] > -EPSILON);
            if (lm_wgt[k] < 0)
                lm_wgt[k] = 0.0;
        }
    }

    delete[] ndx; delete[] val;
    lpx_delete_prob(lp);

    REAL err = 0;
    for (UINT i = 0; i < n_samples; ++i) {
        err += (get_output(i)[0]*ptd->y(i)[0] <= 0);
    }
    return err / n_samples;
}

} // namespace lemga

---