seqParallel: Apply Functions in Parallel
In zhengxwen/SeqArray: Data Management of Large-Scale Whole-Genome Sequence Variant Calls

seqParallel

R Documentation

Apply Functions in Parallel

Description

Applies a user-defined function in parallel.

Usage

seqParallel(cl=seqGetParallel(), gdsfile, FUN,
    split=c("by.variant", "by.sample", "none"), .combine="unlist",
    .selection.flag=FALSE, .initialize=NULL, .finalize=NULL, .initparam=NULL,
    .balancing=FALSE, .bl_size=10000L, .bl_progress=FALSE, ...)
seqParApply(cl=seqGetParallel(), x, FUN, load.balancing=TRUE, ...)

Arguments

`cl`	`NULL` or `FALSE`: serial processing; `TRUE`: multicore processing (the maximum number of cores minor one); a numeric value: the number of cores to be used; a `cluster` object for parallel processing, created by the functions in the package `parallel`, like `makeCluster`; a `BiocParallelParam` object from the BiocParallel package. See details
`gdsfile`	a `SeqVarGDSClass` object, or `NULL`
`FUN`	the function to be applied, should be like `FUN(gdsfile, ...)` if `gdsfile` is given, or `FUN(...)` if `gdsfile=NULL`
`split`	split the dataset by variant or sample according to multiple processes, or "none" for no split; `split="by.variant"` by default
`.combine`	define a fucntion for combining results from different processes; by default, `"unlist"` is used, to produce a vector which contains all the atomic components, via `unlist(..., recursive=FALSE)`; `"list"`, return a list of results created by child processes; `"none"`, no return; or a function with one or two arguments, like "+"
`.selection.flag`	`TRUE` – passes a logical vector of selection to the second argument of `FUN(gdsfile, selection, ...)`
`.initialize`	a user-defined function for initializing workers, should have two arguments (process_id, param)
`.finalize`	a user-defined function for finalizing workers, should have two arguments (process_id, param)
`.initparam`	parameters passed to `.initialize` and `.initialize`
`.balancing`	load balancing if `TRUE`
`.bl_size`	chuck size, the increment for load balancing, 10000 for variants; only applicable if `.balancing=TRUE`
`.bl_progress`	if `TRUE` and `.balancing=TRUE`, show progress information
`x`	a vector (atomic or list), passed to `FUN`
`load.balancing`	if `TRUE`, call `clusterApplyLB` instead of `clusterApply`
`...`	optional arguments to `FUN`

Details

When cl is TRUE or a numeric value, forking techniques are used to create a new child process as a copy of the current R process, see ?parallel::mcfork. However, forking is not available on Windows, and makeCluster is called to make a cluster which will be deallocated after calling FUN.

It is strongly suggested to use seqParallel together with seqParallelSetup. seqParallelSetup could work around the problem of forking on Windows, without allocating clusters frequently.

The user-defined function could use two predefined variables SeqArray:::process_count and SeqArray:::process_index to tell the total number of cluster nodes and which cluster node being used.

seqParallel(, gdsfile=NULL, FUN=..., split="none") could be used to setup multiple streams of pseudo-random numbers, and see nextRNGStream or nextRNGSubStream in the package parallel.

Value

A vector or list of values.

Author(s)

Xiuwen Zheng

Examples

library(parallel)

# choose an appropriate cluster size or number of cores
seqParallelSetup(2)


# the GDS file
(gds.fn <- seqExampleFileName("gds"))

# display
(gdsfile <- seqOpen(gds.fn))

# the uniprocessor version
afreq1 <- seqParallel(, gdsfile, FUN = function(f) {
        seqApply(f, "genotype", as.is="double",
            FUN=function(x) mean(x==0, na.rm=TRUE))
    }, split="by.variant")

length(afreq1)
summary(afreq1)


# run in parallel
afreq2 <- seqParallel(, gdsfile, FUN = function(f) {
        seqApply(f, "genotype", as.is="double",
            FUN=function(x) mean(x==0, na.rm=TRUE))
    }, split="by.variant")

length(afreq2)
summary(afreq2)


# check
length(afreq1)  # 1348
all(afreq1 == afreq2)

################################################################
# check -- variant splits

seqParallel(, gdsfile, FUN = function(f) {
        v <- seqGetFilter(f)
        sum(v$variant.sel)
    }, split="by.variant")
# [1] 674 674


################################################################

seqParallel(, NULL, FUN = function() {
        paste(SeqArray:::process_index, SeqArray:::process_count, sep=" / ")
    }, split="none")

seqParallel(, NULL, FUN = function() {
        SeqArray:::process_index
    }, split="none", .combine=function(i) print(i))

seqParallel(, NULL, FUN = function() {
        SeqArray:::process_index
    }, split="none", .combine="+")


################################################################


# close the GDS file
seqClose(gdsfile)

# clear the parallel cluster
seqParallelSetup(FALSE)

zhengxwen/SeqArray documentation built on Nov. 19, 2024, 1:04 p.m.