marginal.h File Reference

Go to the source code of this file.

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

Function Documentation

marginal_setup()

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

Definition at line 351 of file marginal.cc.

References VW::config::option_group_definition::add(), VW::config::options_i::add_and_parse(), LEARNER::as_singleline(), LEARNER::init_learner(), LEARNER::make_base(), VW::config::make_option(), MARGINAL::save_load(), setup_base(), and VW::config::options_i::was_supplied().

Referenced by parse_reductions().

{
  free_ptr<MARGINAL::data> d = scoped_calloc_or_throw<MARGINAL::data>();
  std::string marginal;

  option_group_definition marginal_options("VW options");
  marginal_options.add(make_option("marginal", marginal).keep().help("substitute marginal label estimates for ids"));
  marginal_options.add(
      make_option("initial_denominator", d->initial_denominator).default_value(1.f).help("initial denominator"));
  marginal_options.add(
      make_option("initial_numerator", d->initial_numerator).default_value(0.5f).help("initial numerator"));
  marginal_options.add(make_option("compete", d->compete).help("enable competition with marginal features"));
  marginal_options.add(
      make_option("update_before_learn", d->update_before_learn).help("update marginal values before learning"));
  marginal_options.add(make_option("unweighted_marginals", d->unweighted_marginals)
                           .help("ignore importance weights when computing marginals"));
  marginal_options.add(
      make_option("decay", d->decay).default_value(0.f).help("decay multiplier per event (1e-3 for example)"));
  options.add_and_parse(marginal_options);

  if (!options.was_supplied("marginal"))
  {
    return nullptr;
  }

  d->all = &all;

  for (size_t u = 0; u < 256; u++)
    if (marginal.find((char)u) != std::string::npos)
      d->id_features[u] = true;

  LEARNER::learner<MARGINAL::data, example>& ret =
      init_learner(d, as_singleline(setup_base(options, all)), predict_or_learn<true>, predict_or_learn<false>);
  ret.set_save_load(save_load);

  return make_base(ret);
}