The combination of the machine learning models for molecular energies and atomic forces. More...

Collaboration diagram for Scine::Swoose::MachineLearning::MolecularMachineLearningModel:

Public Member Functions
	MolecularMachineLearningModel ()=default
	Default constructor.

void	setReferenceData (const Utils::MolecularTrajectory &structures, const std::vector< ForcesCollection > &refForces)
	Sets the reference data. More...

void	trainEnergyModel ()
	Trains the energy model.

void	trainForcesModel ()
	Trains the forces model.

double	predictEnergy (const Utils::AtomCollection &structure)
	Predicts the energy for a given structure.

ForcesCollection	predictForces (const Utils::AtomCollection &structure)
	Predicts the forces for a given structure.

std::pair< double, double >	evaluateEnergyModel (int k)
	Validates the energy model via k-fold cross validation. More...

std::pair< double, double >	evaluateForcesModel (int k, bool pooledVariance=true)
	Validates the model for the atomic forces via k-fold cross validation. More...

Utils::MachineLearning::KernelRidgeRegression &	energyPredictor ()
	Accessor for the underlying energy model.

Utils::MachineLearning::KernelRidgeRegression &	forcePredictor (int atomIndex)
	Accessor for the underlying force model of the atom with index 'atomIndex'.

Detailed Description

The combination of the machine learning models for molecular energies and atomic forces.

Member Function Documentation

std::pair< double, double > Scine::Swoose::MachineLearning::MolecularMachineLearningModel::evaluateEnergyModel ( int k )

Validates the energy model via k-fold cross validation.

Parameters

k	The number of subsets k for the k-fold cross validation algorithm.

Returns: The mean absolute error and the standard deviation of the model evaluation as a std::pair.

std::pair< double, double > Scine::Swoose::MachineLearning::MolecularMachineLearningModel::evaluateForcesModel	(	int	k,
		bool	pooledVariance = `true`
	)

Validates the model for the atomic forces via k-fold cross validation.

Parameters

k	The number of subsets k for the k-fold cross validation algorithm.
pooledVariance	Decides whether the standard deviation estimate for the forces is based on the principle of pooled variance. The alternative is a true combined variance that considers the different MAE means of the atomic forces.

Returns: The combined mean absolute error and the combined standard deviation of the model evaluation for the individual atomic forces as a std::pair.

void Scine::Swoose::MachineLearning::MolecularMachineLearningModel::setReferenceData	(	const Utils::MolecularTrajectory &	structures,
		const std::vector< ForcesCollection > &	refForces
	)

Sets the reference data.

Parameters

structures	The molecular trajectory containing the structures along with their reference energies.
refForces	A vector of the reference forces for each structure.

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