Merge pull request #491 from imperialCHEPI/time_trend

Time trend for expected values

src/HealthGPS.Input/model_parser.cpp

@@ -24,29 +24,30 @@
 
 namespace hgps::input {
 
-hgps::RiskFactorSexAgeTable load_risk_factor_expected(const Configuration &config) {
+std::unique_ptr<hgps::RiskFactorSexAgeTable>
+load_risk_factor_expected(const Configuration &config) {
  MEASURE_FUNCTION();
 
  const auto &info = config.modelling.baseline_adjustment;
  if (!hgps::core::case_insensitive::equals(info.format, "CSV")) {
  throw hgps::core::HgpsException{"Unsupported file format: " + info.format};
  }
 
- auto table = hgps::RiskFactorSexAgeTable{};
+ auto table = std::make_unique<hgps::RiskFactorSexAgeTable>();
  const auto male_filename = info.file_names.at("factorsmean_male").string();
  const auto female_filename = info.file_names.at("factorsmean_female").string();
  try {
- table.emplace_row(hgps::core::Gender::male,
- load_baseline_from_csv(male_filename, info.delimiter));
- table.emplace_row(hgps::core::Gender::female,
- load_baseline_from_csv(female_filename, info.delimiter));
+ table->emplace_row(hgps::core::Gender::male,
+  load_baseline_from_csv(male_filename, info.delimiter));
+ table->emplace_row(hgps::core::Gender::female,
+  load_baseline_from_csv(female_filename, info.delimiter));
  } catch (const std::runtime_error &ex) {
  throw hgps::core::HgpsException{fmt::format("Failed to parse adjustment file: {} or {}. {}",
  male_filename, female_filename, ex.what())};
  }
 
  const auto max_age = static_cast<std::size_t>(config.settings.age_range.upper());
- for (const auto &sex : table) {
+ for (const auto &sex : *table) {
  for (const auto &factor : sex.second) {
  if (factor.second.size() <= max_age) {
  throw hgps::core::HgpsException{
@@ -177,13 +178,14 @@ load_staticlinear_risk_model_definition(const nlohmann::json &opt, const Configu
  policy_covariance_table.num_columns()};
 
  // Risk factor and intervention policy: names, models, parameters and correlation/covariance.
- std::vector<hgps::core::Identifier> names;
- std::vector<hgps::LinearModelParams> models;
- std::vector<hgps::core::DoubleInterval> ranges;
+ std::vector<core::Identifier> names;
+ std::vector<LinearModelParams> models;
+ std::vector<core::DoubleInterval> ranges;
  std::vector<double> lambda;
  std::vector<double> stddev;
- std::vector<hgps::LinearModelParams> policy_models;
- std::vector<hgps::core::DoubleInterval> policy_ranges;
+ std::vector<LinearModelParams> policy_models;
+ std::vector<core::DoubleInterval> policy_ranges;
+ auto expected_trend = std::make_unique<std::unordered_map<core::Identifier, double>>();
 
  size_t i = 0;
  for (const auto &[key, json_params] : opt["RiskFactorModels"].items()) {
@@ -240,6 +242,9 @@ load_staticlinear_risk_model_definition(const nlohmann::json &opt, const Configu
  std::any_cast<double>(policy_covariance_table.column(i).value(j));
  }
 
+ // Load expected value trends.
+ (*expected_trend)[key] = json_params["ExpectedTrend"].get<double>();
+
  // Increment table column index.
  i++;
  }
@@ -268,15 +273,15 @@ load_staticlinear_risk_model_definition(const nlohmann::json &opt, const Configu
  Eigen::MatrixXd{Eigen::LLT<Eigen::MatrixXd>{policy_covariance}.matrixL()};
 
  // Risk factor expected values by sex and age.
- hgps::RiskFactorSexAgeTable expected = load_risk_factor_expected(config);
+ auto expected = load_risk_factor_expected(config);
 
  // Check expected values are defined for all risk factors.
  for (const auto &name : names) {
- if (!expected.at(hgps::core::Gender::male).contains(name)) {
+ if (!expected->at(hgps::core::Gender::male).contains(name)) {
  throw hgps::core::HgpsException{fmt::format(
  "'{}' is not defined in male risk factor expected values.", name.to_string())};
  }
- if (!expected.at(hgps::core::Gender::female).contains(name)) {
+ if (!expected->at(hgps::core::Gender::female).contains(name)) {
  throw hgps::core::HgpsException{fmt::format(
  "'{}' is not defined in female risk factor expected values.", name.to_string())};
  }
@@ -327,9 +332,9 @@ load_staticlinear_risk_model_definition(const nlohmann::json &opt, const Configu
  const double physical_activity_stddev = opt["PhysicalActivityStdDev"].get<double>();
 
  return std::make_unique<hgps::StaticLinearModelDefinition>(
- std::move(expected), std::move(names), std::move(models), std::move(ranges),
- std::move(lambda), std::move(stddev), std::move(cholesky), std::move(policy_models),
- std::move(policy_ranges), std::move(policy_cholesky), info_speed,
+ std::move(expected), std::move(expected_trend), std::move(names), std::move(models),
+ std::move(ranges), std::move(lambda), std::move(stddev), std::move(cholesky),
+ std::move(policy_models), std::move(policy_ranges), std::move(policy_cholesky), info_speed,
  std::move(rural_prevalence), std::move(income_models), physical_activity_stddev);
 }
 
@@ -425,10 +430,12 @@ load_ebhlm_risk_model_definition(const nlohmann::json &opt, const Configuration
  }
 
  // Risk factor expected values by sex and age.
- hgps::RiskFactorSexAgeTable expected = load_risk_factor_expected(config);
+ auto expected = load_risk_factor_expected(config);
+ auto expected_trend = std::make_unique<std::unordered_map<core::Identifier, double>>();
 
  return std::make_unique<hgps::DynamicHierarchicalLinearModelDefinition>(
- std::move(expected), std::move(equations), std::move(variables), percentage);
+ std::move(expected), std::move(expected_trend), std::move(equations), std::move(variables),
+ percentage);
 }
 // NOLINTEND(readability-function-cognitive-complexity)
 
@@ -437,7 +444,8 @@ load_kevinhall_risk_model_definition(const nlohmann::json &opt, const Configurat
  MEASURE_FUNCTION();
 
  // Risk factor expected values by sex and age.
- hgps::RiskFactorSexAgeTable expected = load_risk_factor_expected(config);
+ auto expected = load_risk_factor_expected(config);
+ auto expected_trend = std::make_unique<std::unordered_map<core::Identifier, double>>();
 
  // Nutrient groups.
  std::unordered_map<hgps::core::Identifier, double> energy_equation;
@@ -507,9 +515,10 @@ load_kevinhall_risk_model_definition(const nlohmann::json &opt, const Configurat
  {hgps::core::Gender::male, opt["HeightSlope"]["Male"].get<double>()}};
 
  return std::make_unique<hgps::KevinHallModelDefinition>(
- std::move(expected), std::move(energy_equation), std::move(nutrient_ranges),
- std::move(nutrient_equations), std::move(food_prices), std::move(weight_quantiles),
- std::move(epa_quantiles), std::move(height_stddev), std::move(height_slope));
+ std::move(expected), std::move(expected_trend), std::move(energy_equation),
+ std::move(nutrient_ranges), std::move(nutrient_equations), std::move(food_prices),
+ std::move(weight_quantiles), std::move(epa_quantiles), std::move(height_stddev),
+ std::move(height_slope));
 }
 
 std::pair<hgps::RiskFactorModelType, std::unique_ptr<hgps::RiskFactorModelDefinition>>

src/HealthGPS.Input/model_parser.h

@@ -17,7 +17,7 @@ namespace hgps::input {
 /// @brief Loads risk factor expected values from a file
 /// @param config The model configuration
 /// @return An instance of the hgps::RiskFactorSexAgeTable type
-hgps::RiskFactorSexAgeTable load_risk_factor_expected(const Configuration &config);
+std::unique_ptr<hgps::RiskFactorSexAgeTable> load_risk_factor_expected(const Configuration &config);
 
 /// @brief Loads either a static or dynamic dummy risk factor model from a JSON file
 /// @param type Model type (static or dynamic)

src/HealthGPS/dynamic_hierarchical_linear_model.cpp

@@ -3,16 +3,18 @@
 
 #include "HealthGPS.Core/exception.h"
 
+#include <utility>
 #include <vector>
 
 namespace hgps {
 
 DynamicHierarchicalLinearModel::DynamicHierarchicalLinearModel(
- const RiskFactorSexAgeTable &expected,
+ std::shared_ptr<RiskFactorSexAgeTable> expected,
+ std::shared_ptr<std::unordered_map<core::Identifier, double>> expected_trend,
  const std::map<core::IntegerInterval, AgeGroupGenderEquation> &equations,
  const std::map<core::Identifier, core::Identifier> &variables, double boundary_percentage)
- : RiskFactorAdjustableModel{expected}, equations_{equations}, variables_{variables},
- boundary_percentage_{boundary_percentage} {}
+ : RiskFactorAdjustableModel{std::move(expected), std::move(expected_trend)},
+ equations_{equations}, variables_{variables}, boundary_percentage_{boundary_percentage} {}
 
 RiskFactorModelType DynamicHierarchicalLinearModel::type() const noexcept {
  return RiskFactorModelType::Dynamic;
@@ -137,11 +139,13 @@ double DynamicHierarchicalLinearModel::sample_normal_with_boundary(Random &rando
 }
 
 DynamicHierarchicalLinearModelDefinition::DynamicHierarchicalLinearModelDefinition(
- RiskFactorSexAgeTable expected,
+ std::unique_ptr<RiskFactorSexAgeTable> expected,
+ std::unique_ptr<std::unordered_map<core::Identifier, double>> expected_trend,
  std::map<core::IntegerInterval, AgeGroupGenderEquation> equations,
  std::map<core::Identifier, core::Identifier> variables, const double boundary_percentage)
- : RiskFactorAdjustableModelDefinition{std::move(expected)}, equations_{std::move(equations)},
- variables_{std::move(variables)}, boundary_percentage_{boundary_percentage} {
+ : RiskFactorAdjustableModelDefinition{std::move(expected), std::move(expected_trend)},
+ equations_{std::move(equations)}, variables_{std::move(variables)},
+ boundary_percentage_{boundary_percentage} {
 
  if (equations_.empty()) {
  throw core::HgpsException("The model equations definition must not be empty");
@@ -152,9 +156,8 @@ DynamicHierarchicalLinearModelDefinition::DynamicHierarchicalLinearModelDefiniti
 }
 
 std::unique_ptr<RiskFactorModel> DynamicHierarchicalLinearModelDefinition::create_model() const {
- const auto &expected = get_risk_factor_expected();
- return std::make_unique<DynamicHierarchicalLinearModel>(expected, equations_, variables_,
- boundary_percentage_);
+ return std::make_unique<DynamicHierarchicalLinearModel>(expected_, expected_trend_, equations_,
+ variables_, boundary_percentage_);
 }
 
 } // namespace hgps

src/HealthGPS/dynamic_hierarchical_linear_model.h

@@ -39,11 +39,13 @@ class DynamicHierarchicalLinearModel final : public RiskFactorAdjustableModel {
  public:
  /// @brief Initialises a new instance of the DynamicHierarchicalLinearModel class
  /// @param expected The expected values
+ /// @param expected_trend The expected trend of risk factor values
  /// @param equations The linear regression equations
  /// @param variables The factors delta variables mapping
  /// @param boundary_percentage The boundary percentage to sample
  DynamicHierarchicalLinearModel(
- const RiskFactorSexAgeTable &expected,
+ std::shared_ptr<RiskFactorSexAgeTable> expected,
+ std::shared_ptr<std::unordered_map<core::Identifier, double>> expected_trend,
  const std::map<core::IntegerInterval, AgeGroupGenderEquation> &equations,
  const std::map<core::Identifier, core::Identifier> &variables,
  const double boundary_percentage);
@@ -80,12 +82,14 @@ class DynamicHierarchicalLinearModelDefinition : public RiskFactorAdjustableMode
  public:
  /// @brief Initialises a new instance of the DynamicHierarchicalLinearModelDefinition class
  /// @param expected The expected values
+ /// @param expected_trend The expected trend of risk factor values
  /// @param equations The linear regression equations
  /// @param variables The factors delta variables mapping
  /// @param boundary_percentage The boundary percentage to sample
  /// @throws std::invalid_argument for empty model equations definition
  DynamicHierarchicalLinearModelDefinition(
- RiskFactorSexAgeTable expected,
+ std::unique_ptr<RiskFactorSexAgeTable> expected,
+ std::unique_ptr<std::unordered_map<core::Identifier, double>> expected_trend,
  std::map<core::IntegerInterval, AgeGroupGenderEquation> equations,
  std::map<core::Identifier, core::Identifier> variables,
  const double boundary_percentage = 0.05);