pmclust (version 0.1-2)

assign.N.sample: Obtain a Set of Random Samples for X.spmd

Description

This utility function samples data randomly from X.spmd to form a relatively small subset of original data. The EM algorithm on the smaller subset is topically performing fast and capturing rough structures of entire dataset.

Usage

assign.N.sample(total.sample = 5000, N.org.spmd)

Arguments

total.sample

a total number of samples which will be selected from the original data X.spmd.

N.org.spmd

the original data size, i.e. nrow(X.spmd).

Value

• A list variable will be returned and containing: ll{N total sample size across all $S$ processors N.spmd sample size of given processor N.allspmds a collection of sample sizes for all $S$ processors ID.spmd index of selected samples ranged from 1 to N.org.spmd } Note that N and N.allspmds are the same across all $S$ processors, but N.spmd and ID.spmd are most likely all distinct. The lengths of these elements are $1$ for N and N.spmd, $S$ for N.allspmd, and N.spmd for ID.spmd.

Details

This utility function performs simple random sampling without replacement for the original dataset X.spmd. Different random seeds should be set before calling this function. A easy way is to call for example set.seed(123 + mpi.comm.rank()).

References

High Performance Statistical Computing Website: http://thirteen-01.stat.iastate.edu/snoweye/hpsc/

See Also

set.global

Examples

# Save code in a file "demo.r" and run in 4 processors by
# > mpirun -np 4 Rscript demo.r

### Setup mpi environment.
library(Rmpi)
invisible(mpi.comm.dup(0, 1))

### Generate an example data.
set.seed(123 + mpi.comm.rank())
N.org.spmd <- 5000 + sample(1:1000, 1)
ret.spmd <- assign.N.sample(total.sample = 5000, N.org.spmd)
cat("Rank:", mpi.comm.rank(), " Size:", ret.spmd$N.spmd,
    "\n", sep = "")

### Quit Rmpi.
mpi.quit()