cs_active.h File Reference

Go to the source code of this file.

Functions
LEARNER::base_learner *	cs_active_setup (VW::config::options_i &options, vw &all)

Function Documentation

cs_active_setup()

LEARNER::base_learner* cs_active_setup	(	VW::config::options_i &	options,
		vw &	all
	)

Definition at line 310 of file cs_active.cc.

References VW::config::option_group_definition::add(), add(), VW::config::options_i::add_and_parse(), LEARNER::as_singleline(), vw::cost_sensitive, COST_SENSITIVE::cs_label, f, finish_example(), loss_function::getType(), LEARNER::init_learner(), vw::loss, parser::lp, LEARNER::make_base(), VW::config::make_option(), prediction_type::multilabels, vw::p, vw::sd, LEARNER::learner< T, E >::set_finish_example(), vw::set_minmax, setup_base(), THROW, vw::trace_message, and VW::config::options_i::was_supplied().

Referenced by parse_reductions().

{
  auto data = scoped_calloc_or_throw<cs_active>();

  bool simulation = false;
  int domination;
  option_group_definition new_options("Cost-sensitive Active Learning");
  new_options
      .add(make_option("cs_active", data->num_classes).keep().help("Cost-sensitive active learning with <k> costs"))
      .add(make_option("simulation", simulation).help("cost-sensitive active learning simulation mode"))
      .add(make_option("baseline", data->is_baseline).help("cost-sensitive active learning baseline"))
      .add(make_option("domination", domination)
               .default_value(1)
               .help("cost-sensitive active learning use domination. Default 1"))
      .add(make_option("mellowness", data->c0).default_value(0.1f).help("mellowness parameter c_0. Default 0.1."))
      .add(make_option("range_c", data->c1)
               .default_value(0.5f)
               .help("parameter controlling the threshold for per-label cost uncertainty. Default 0.5."))
      .add(make_option("max_labels", data->max_labels).default_value(-1).help("maximum number of label queries."))
      .add(make_option("min_labels", data->min_labels).default_value(-1).help("minimum number of label queries."))
      .add(make_option("cost_max", data->cost_max).default_value(1.f).help("cost upper bound. Default 1."))
      .add(make_option("cost_min", data->cost_min).default_value(0.f).help("cost lower bound. Default 0."))
      // TODO replace with trace and quiet
      .add(make_option("csa_debug", data->print_debug_stuff).help("print debug stuff for cs_active"));
  options.add_and_parse(new_options);

  data->use_domination = true;
  if (options.was_supplied("domination") && !domination)
    data->use_domination = false;

  if (!options.was_supplied("cs_active"))
    return nullptr;

  data->all = &all;
  data->t = 1;

  auto loss_function_type = all.loss->getType();
  if (loss_function_type != "squared")
    THROW("error: you can't use non-squared loss with cs_active");

  if (options.was_supplied("lda"))
    THROW("error: you can't combine lda and active learning");

  if (options.was_supplied("active"))
    THROW("error: you can't use --cs_active and --active at the same time");

  if (options.was_supplied("active_cover"))
    THROW("error: you can't use --cs_active and --active_cover at the same time");

  if (options.was_supplied("csoaa"))
    THROW("error: you can't use --cs_active and --csoaa at the same time");

  if (!options.was_supplied("adax"))
    all.trace_message << "WARNING: --cs_active should be used with --adax" << endl;

  all.p->lp = cs_label;  // assigning the label parser
  all.set_minmax(all.sd, data->cost_max);
  all.set_minmax(all.sd, data->cost_min);
  for (uint32_t i = 0; i < data->num_classes + 1; i++) data->examples_by_queries.push_back(0);

  learner<cs_active, example>& l = simulation
      ? init_learner(data, as_singleline(setup_base(options, all)), predict_or_learn<true, true>,
            predict_or_learn<false, true>, data->num_classes, prediction_type::multilabels)
      : init_learner(data, as_singleline(setup_base(options, all)), predict_or_learn<true, false>,
            predict_or_learn<false, false>, data->num_classes, prediction_type::multilabels);

  l.set_finish_example(finish_example);
  base_learner* b = make_base(l);
  all.cost_sensitive = b;
  return b;
}