Inheritance diagram for hmlp::mpigofmm::randomsplit< SPDMATRIX, N_SPLIT, T >:

Public Member Functions
	randomsplit (SPDMATRIX &K)

vector< vector< size_t > >	operator() (vector< size_t > &gids) const

vector< vector< size_t > >	operator() (vector< size_t > &gids, mpi::Comm comm) const

Public Member Functions inherited from hmlp::gofmm::randomsplit< SPDMATRIX, N_SPLIT, T >
	randomsplit (SPDMATRIX &K)

vector< vector< size_t > >	operator() (vector< size_t > &gids) const

Additional Inherited Members
Public Attributes inherited from hmlp::gofmm::randomsplit< SPDMATRIX, N_SPLIT, T >
SPDMATRIX *	Kptr = NULL

DistanceMetric	metric = ANGLE_DISTANCE

Detailed Description

end struct centersplit

Member Function Documentation

template<typename SPDMATRIX , int N_SPLIT, typename T >

vector<vector<size_t> > hmlp::mpigofmm::randomsplit< SPDMATRIX, N_SPLIT, T >::operator() ( vector< size_t > & gids ) const

inline

Shared-memory operator.

template<typename SPDMATRIX , int N_SPLIT, typename T >

vector<vector<size_t> > hmlp::mpigofmm::randomsplit< SPDMATRIX, N_SPLIT, T >::operator()	(	vector< size_t > &	gids,
		mpi::Comm	comm
	)		const

inline

Distributed operator.

All assertions

Declaration

Reduce to get the total size of gids.

n = sum( num_points_owned ) over all MPI processes in comm

Early return

Randomly select two points p and q

Create a pair <gids.size(), rank> for MPI Allreduce

max_pair = max( local_max_pairs ) over all MPI processes in comm

Bcast gidf2c from the rank that has the most gids

Choose the second MPI rank

max_pair = max( local_max_pairs ) over all MPI processes in comm

Bcast gidf2c from the rank that has the most gids

We use relative distances (dip - diq) for clustering.

Split gids into two clusters using median split.

Perform P2P redistribution.

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