oaa.h File Reference

Go to the source code of this file.

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

Function Documentation

oaa_setup()

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

Definition at line 203 of file oaa.cc.

References VW::config::option_group_definition::add(), add(), VW::config::options_i::add_and_parse(), LEARNER::as_singleline(), vw::delete_prediction, delete_scalars(), vw::get_random_state(), namedlabels::getK(), loss_function::getType(), LEARNER::init_multiclass_learner(), shared_data::ldict, learn_randomized(), vw::loss, LEARNER::make_base(), VW::config::make_option(), prediction_type::multiclass, oaa::num_subsample, vw::p, vw::raw_prediction, shared_data::report_multiclass_log_loss, prediction_type::scalars, vw::sd, LEARNER::learner< T, E >::set_finish_example(), LEARNER::learner< T, E >::set_learn(), setup_base(), THROW, vw::trace_message, and VW::config::options_i::was_supplied().

Referenced by parse_reductions().

{
  auto data = scoped_calloc_or_throw<oaa>();
  bool probabilities = false;
  bool scores = false;
  option_group_definition new_options("One Against All Options");
  new_options.add(make_option("oaa", data->k).keep().help("One-against-all multiclass with <k> labels"))
      .add(make_option("oaa_subsample", data->num_subsample)
               .help("subsample this number of negative examples when learning"))
      .add(make_option("probabilities", probabilities).help("predict probabilites of all classes"))
      .add(make_option("scores", scores).help("output raw scores per class"));
  options.add_and_parse(new_options);

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

  if (all.sd->ldict && (data->k != all.sd->ldict->getK()))
    THROW("error: you have " << all.sd->ldict->getK() << " named labels; use that as the argument to oaa")

  data->all = &all;
  data->pred = calloc_or_throw<polyprediction>(data->k);
  data->subsample_order = nullptr;
  data->subsample_id = 0;
  if (data->num_subsample > 0)
  {
    if (data->num_subsample >= data->k)
    {
      data->num_subsample = 0;
      all.trace_message << "oaa is turning off subsampling because your parameter >= K" << std::endl;
    }
    else
    {
      data->subsample_order = calloc_or_throw<uint32_t>(data->k);
      for (size_t i = 0; i < data->k; i++) data->subsample_order[i] = (uint32_t)i;
      for (size_t i = 0; i < data->k; i++)
      {
        size_t j = (size_t)(all.get_random_state()->get_and_update_random() * (float)(data->k - i)) + i;
        uint32_t tmp = data->subsample_order[i];
        data->subsample_order[i] = data->subsample_order[j];
        data->subsample_order[j] = tmp;
      }
    }
  }

  oaa* data_ptr = data.get();
  LEARNER::learner<oaa, example>* l;
  auto base = as_singleline(setup_base(options, all));
  if (probabilities || scores)
  {
    all.delete_prediction = delete_scalars;
    if (probabilities)
    {
      auto loss_function_type = all.loss->getType();
      if (loss_function_type != "logistic")
        all.trace_message << "WARNING: --probabilities should be used only with --loss_function=logistic" << std::endl;
      // the three boolean template parameters are: is_learn, print_all and scores
      l = &LEARNER::init_multiclass_learner(data, base, predict_or_learn<true, false, true, true>,
          predict_or_learn<false, false, true, true>, all.p, data->k, prediction_type::scalars);
      all.sd->report_multiclass_log_loss = true;
      l->set_finish_example(finish_example_scores<true>);
    }
    else
    {
      l = &LEARNER::init_multiclass_learner(data, base, predict_or_learn<true, false, true, false>,
          predict_or_learn<false, false, true, false>, all.p, data->k, prediction_type::scalars);
      l->set_finish_example(finish_example_scores<false>);
    }
  }
  else if (all.raw_prediction > 0)
    l = &LEARNER::init_multiclass_learner(data, base, predict_or_learn<true, true, false, false>,
        predict_or_learn<false, true, false, false>, all.p, data->k, prediction_type::multiclass);
  else
    l = &LEARNER::init_multiclass_learner(data, base, predict_or_learn<true, false, false, false>,
        predict_or_learn<false, false, false, false>, all.p, data->k, prediction_type::multiclass);

  if (data_ptr->num_subsample > 0)
  {
    l->set_learn(learn_randomized);
    l->set_finish_example(MULTICLASS::finish_example_without_loss<oaa>);
  }

  return make_base(*l);
}