Method majorChooseRow

highs/simplex/HEkkDualMulti.cpp:67–171

}

void HEkkDual::majorChooseRow() {
  /**
   * 0. Initial check to see if we need to do it again
   */
  if (ekk_instance_.info_.update_count == 0) multi_chooseAgain = 1;
  if (!multi_chooseAgain) return;
  multi_chooseAgain = 0;
  multi_iteration++;
  /**
   * Major loop:
   *     repeat 1-5, until we found a good sets of choices
   */
  std::vector<HighsInt> choiceIndex(multi_num, 0);
  std::vector<double>& edge_weight = ekk_instance_.dual_edge_weight_;
  for (;;) {
    // 1. Multiple CHUZR
    HighsInt initialCount = 0;

    // Call the hyper-graph method, but partSwitch=0 so just uses
    // choose_multi_global
    dualRHS.chooseMultiHyperGraphAuto(choiceIndex.data(), &initialCount,
                                      multi_num);
    // dualRHS.chooseMultiGlobal(choiceIndex.data(), &initialCount, multi_num);
    if (initialCount == 0 && dualRHS.workCutoff == 0) {
      // OPTIMAL
      return;
    }

    // 2. Shrink the size by cutoff
    HighsInt choiceCount = 0;
    for (HighsInt i = 0; i < initialCount; i++) {
      HighsInt iRow = choiceIndex[i];
      if (dualRHS.work_infeasibility[iRow] / edge_weight[iRow] >=
          dualRHS.workCutoff) {
        choiceIndex[choiceCount++] = iRow;
      }
    }

    if (initialCount == 0 || choiceCount <= initialCount / 3) {
      // Need to do the list again
      dualRHS.createInfeasList(ekk_instance_.info_.col_aq_density);
      continue;
    }

    // 3. Store the choiceIndex to buffer
    for (HighsInt ich = 0; ich < multi_num; ich++)
      multi_choice[ich].row_out = kNoRowChosen;
    for (HighsInt ich = 0; ich < choiceCount; ich++)
      multi_choice[ich].row_out = choiceIndex[ich];

    // 4. Parallel BTRAN and compute weight
    majorChooseRowBtran();

    // 5. Update row densities
    for (HighsInt ich = 0; ich < multi_num; ich++) {
      if (multi_choice[ich].row_out >= 0) {
        const double local_row_ep_density =
            (double)multi_choice[ich].row_ep.count / solver_num_row;