Grid search tuning strategy for model hyperparameter optimisation. More...

#include <GridSearchTuner.h>

Inheritance diagram for lucid::GridSearchTuner:

Public Member Functions
	GridSearchTuner (std::vector< ParameterValues > parameters, std::size_t n_jobs=0)
	Construct a new GridSearchTuner object with the given `parameters`.

std::size_t	n_jobs () const
	Get read-only access to the number of parallel jobs of the grid search.

const std::vector< ParameterValues > &	parameters () const
	Get read-only access to the parameters of the grid search.

std::string	to_string () const override
	Obtain the string representation of this object.

Public Member Functions inherited from lucid::Tuner
void	tune (Estimator &estimator, ConstMatrixRef training_inputs, ConstMatrixRef training_outputs) const
	Optimise the kernel hyperparameters.

void	tune_online (Estimator &estimator, ConstMatrixRef training_inputs, const OutputComputer &training_outputs) const
	Optimise the kernel hyperparameters.

Private Member Functions
void	tune_impl (Estimator &estimator, ConstMatrixRef training_inputs, const OutputComputer &training_outputs) const override
	Optimise the kernel hyperparameters.

Private Attributes
std::size_t	n_jobs_
	Number of parallel jobs to run during tuning.

std::vector< ParameterValues >	parameters_
	List of parameter values to be tuned, with the values to be tested.

std::vector< Index >	parameters_max_indices_
	Maximum indices for each parameter, used to iterate over the grid.

Detailed Description

Grid search tuning strategy for model hyperparameter optimisation.

The GridSearchTuner is responsible for iterating over a predefined grid of hyperparameter combinations to optimise the performance of a given estimator. More precisely, the Estimator::score method is called for each combination of hyperparameters, keeping track of the best combination (i.e., the one producing the highest value) found during the search. When all combinations have been evaluated, the best hyperparameters are set on the estimator. This tuner is suitable for scenarios where exhaustive search over a finite hyperparameter space is required.

See also: scorer::r2_score

Constructor & Destructor Documentation

◆ GridSearchTuner()

lucid::GridSearchTuner::GridSearchTuner	(	std::vector< ParameterValues >	parameters,
		std::size_t	n_jobs = 0 )

explicit

Construct a new GridSearchTuner object with the given parameters.

Parameters

parameters	parameters to be tuned, with the values to be tested
n_jobs	number of parallel jobs to run during tuning. If set to 0, it defaults to max(1, CPU cores - 2).

Member Function Documentation

◆ n_jobs()

std::size_t lucid::GridSearchTuner::n_jobs ( ) const

inlinenodiscard

Get read-only access to the number of parallel jobs of the grid search.

Returns: number of parallel jobs of the grid search

◆ parameters()

const std::vector< ParameterValues > & lucid::GridSearchTuner::parameters ( ) const

inlinenodiscard

Get read-only access to the parameters of the grid search.

Returns: parameters of the grid search

◆ to_string()

std::string lucid::GridSearchTuner::to_string ( ) const

nodiscardoverridevirtual

Obtain the string representation of this object.

Returns: string representation of this object

Reimplemented from lucid::Tuner.

◆ tune_impl()

void lucid::GridSearchTuner::tune_impl	(	Estimator &	estimator,
		ConstMatrixRef	training_inputs,
		const OutputComputer &	training_outputs ) const

overrideprivatevirtual

Optimise the kernel hyperparameters.

Starting from the initial guess, the optimiser finds the best hyperparameters for the kernel. It is up to the subclass to determine the optimisation algorithm used.

Note: The method must call the consolidate method of the estimator to ensure that the kernel is ready to be used.

Precondition: The number of rows in the training inputs should be equal to the number of rows in the training outputs.

Todo: If the OutputComputer becomes a bottleneck, consider having a separate method for when the outputs are available from the get-go.

Parameters

estimator	estimator to optimise
training_inputs	training input data
training_outputs	training output data. It uses an OutputComputer to compute the outputs when needed

Implements lucid::Tuner.

The documentation for this class was generated from the following files:

lucid/model/GridSearchTuner.h
lucid/model/GridSearchTuner.cpp

Public Member Functions

Private Member Functions

Private Attributes