xval.cpp

for (index_t i=0; i <m_num_runs; ++i)
{
  results[i]=evaluate_one_run();
}

float64_t CCrossValidation::evaluate_one_run()
{
  		m_machine->set_store_model_features(true);
  		for (index_t i=0; i <num_subsets; ++i)
  		{
  		        //train
			m_features->add_subset(inverse_subset_indices);
  		      	m_labels->add_subset(inverse_subset_indices);
        		m_machine->train(m_features);
              
              		m_features->remove_subset();
			m_labels->remove_subset();
			        
			//apply
        		m_features->add_subset(subset_indices);
			m_labels->add_subset(subset_indices);
			CLabels* result_labels=m_machine->apply(m_features);
              		results[i]=m_evaluation_criterion->evaluate(result_labels, m_labels);

  		}
}

OK, so you want to create new shallow copies here (of features, labels, and machine). Assign subsets, assign copied labels and features to copied machine, and run

oh and actually also the evaluation instance

I suggest we parallelise over folds, so the stuff I mentioned should happen within the folds loop.

Also make sure that each folds runs in a separate thread, but you should also parallelise over the runs, so openmp needs to kind of merge all folds