multi_dataset.h

Go to the documentation of this file.

 
#if !defined(ULTRA_MULTI_DATASET_H)
#define      ULTRA_MULTI_DATASET_H
 
#include "utility/assert.h"
#include "utility/misc.h"
 
#include <array>
 
namespace ultra::src
{

/// Data/simulations are categorised into three disjoint sets:
/// - *training* used directly for learning;
/// - *validation* used to control overfitting and to measure performance;
/// - *test* used to estimate generalisation performance.
///
enum class dataset_t {training, validation, test};

template<class DAT> concept DataSet = std::ranges::sized_range<DAT>;


template<DataSet D>
class multi_dataset
{
public:
  /// Constructs an empty multi-dataset.
  ///
  /// Each contained dataset is default-constructed. For typical container
  /// types, this results in empty datasets.
  multi_dataset() = default;
 
  // ---- Data access ----
  [[nodiscard]] const D &selected() const noexcept;
  [[nodiscard]] D &selected() noexcept;
  [[nodiscard]] const D &operator[](dataset_t) const noexcept;
  [[nodiscard]] D &operator[](dataset_t) noexcept;
  void select(dataset_t) noexcept;
 
private:
  // Number of datasets managed by this class.
  static constexpr std::size_t dataset_count = 3;
  static_assert(std::to_underlying(dataset_t::test) + 1 == dataset_count,
                "dataset_t enumerators must be contiguous and zero-based");
 
  // Storage for the datasets, indexed by `dataset_t`.
  std::array<D, dataset_count> datasets_;
 
  // Currently selected dataset.
  dataset_t selected_ {dataset_t::training};
};  // class multi_dataset
 

/// evaluates to `false`.
template<class T> struct is_multi_dataset : std::false_type {};
template<class T> struct is_multi_dataset<multi_dataset<T>> : std::true_type {};

template<class T> inline constexpr bool is_multi_dataset_v =
  is_multi_dataset<std::remove_cvref_t<T>>::value;
 
 
#include "kernel/gp/src/multi_dataset.tcc"
 
}  // namespace ultra::src
 
#endif  // include guard