xx_comm_pairwise: Global Pairwise Evaluations

In RBigData/pbdMPI: R Interface to MPI for HPC Clusters (Programming with Big Data Project)

Global Pairwise Evaluations

Description

This function provides global pairwise evaluations.

Usage

comm.pairwise(X, pairid.gbd = NULL,
    FUN = function(x, y, ...){ return(as.vector(dist(rbind(x, y), ...))) },
    ..., diag = FALSE, symmetric = TRUE, comm = .pbd_env$SPMD.CT$comm)

Arguments

X	a common matrix across ranks, or a gbd matrix. (See details.)
pairid.gbd	a pair-wise id in a gbd format. (See details.)
FUN	a function to be evaluated for given pairs.
...	extra variables for FUN.
diag	if matching the same elements, (i, i) for all i.
symmetric	if matching upper triangular elements. TRUE for i >= j only, otherwise for all (i, j).
comm	a communicator number.

Details

This function evaluates the objective function FUN(X[i,], X[j, ]) (usually distance of two elements) on any given pair (i, j) of a matrix X.

The input X should be in common across all ranks if pairid.gbd is provided, e.g. from comm.pairwise(). i.e. X is exactly the same in every ranks, but pairid.gbd is different and in gbd format indicating the row pair (i, j) should be evaluated. The returning gbd matrix is ordered and indexed by pairid.gbd.

Note that checking consistence of X across all ranks is not implemented within this function since that drops performance and may be not accurate.

The input X should be a gbd format in row major blocks (i.e. X.gbd) if pairid.gbd is NULL. A internal pair indices will be built implicitly for evaluation. The returning gbd matrix is ordered and indexed by X.gbd.

Value

This function returns a common matrix with 3 columns named i, j, and value. Each value is the returned value and computed by FUN(X[i,], X[j,]) where (i, j) is the global index as ordered in a distance matrix for i-th row and j-th columns.

Author(s)

Wei-Chen Chen [e-mail], George Ostrouchov, Drew Schmidt, Pragneshkumar Patel, and Hao Yu.

References

Programming with Big Data in R Website: https://pbdr.org/

See Also

comm.pairwise(), and comm.dist().

Examples

## Not run: 
### Save code in a file "demo.r" and run with 2 processors by
### SHELL> mpiexec -np 2 Rscript demo.r

spmd.code <- "
### Initialize
suppressMessages(library(pbdMPI, quietly = TRUE))

### Examples.
comm.set.seed(123456, diff = FALSE)
X <- matrix(rnorm(10), ncol = 2)
id.matrix <- comm.allpairs(nrow(X))

### Method original.
dist.org <- dist(X)

### Method 1.
dist.common <- comm.pairwise(X, pairid.gbd = id.matrix)

### Method 2.
# if(comm.rank() != 0){
#   X <- matrix(0, nrow = 0, ncol = 4)
# }
X.gbd <- comm.as.gbd(X)    ### The other way.
dist.gbd <- comm.pairwise(X.gbd)

### Verify.
d.org <- as.vector(dist.org)
d.1 <- do.call(\"c\", allgather(dist.common[, 3]))
d.2 <- do.call(\"c\", allgather(dist.gbd[, 3]))
comm.print(all(d.org == d.1))
comm.print(all(d.org == d.2))

### Finish.
finalize()
"
# execmpi(spmd.code, nranks = 2L)

## End(Not run)

RBigData/pbdMPI documentation built on Jan. 31, 2024, 10:34 p.m.