Description

NNPot Module.

Abstract class for a neural network potential model. Inherit from this class to implement a specific neural network backend/framework. Refer to torchmodel.h for an example implementation.

Public Member Functions
virtual void	evaluateModel (gmx_enerdata_t , const ArrayRef< RVec > &, ArrayRef< const int > &, ArrayRef< const std::string > &, ArrayRef< RVec > &, ArrayRef< int > &, matrix , PbcType *)=0
	Evaluate NN model. More...

virtual void	setCommRec (const t_commrec *)=0
	set communication record for possible communication of input/output data between ranks

virtual bool	outputsForces () const =0
	helper function to check if model outputs forces

Member Function Documentation

virtual void gmx::INNPotModel::evaluateModel	(	gmx_enerdata_t *	,
		const ArrayRef< RVec > &	,
		ArrayRef< const int > &	,
		ArrayRef< const std::string > &	,
		ArrayRef< RVec > &	,
		ArrayRef< int > &	,
		matrix *	,
		PbcType *
	)

pure virtual

Evaluate NN model.

Includes the preparation of inputs, calling inference of the model, and retrieval of outputs. Currently supported inputs:

atom positions (ArrayRef<RVec>): atomic positions
atom numbers (ArrayRef<int>): atomic numbers
box (matrix): simulation box vectors
pbc type (PbcType): boolean flags for periodic boundary conditions in x, y, z

Implemented in gmx::TorchModel.

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

src/gromacs/applied_forces/nnpot/nnpotmodel.h

Description

Public Member Functions

Member Function Documentation