pbapply: Adding Progress Bar to '*apply' Functions

pbapplyR Documentation

Adding Progress Bar to '*apply' Functions


Adding progress bar to *apply functions, possibly leveraging parallel processing.


pblapply(X, FUN, ..., cl = NULL)
pbeapply(env, FUN, ..., all.names = FALSE, USE.NAMES = TRUE, cl = NULL)
pbwalk(X, FUN, ..., cl = NULL)

pbapply(X, MARGIN, FUN, ..., simplify = TRUE, cl = NULL)

pbsapply(X, FUN, ..., simplify = TRUE, USE.NAMES = TRUE, cl = NULL)
pbvapply(X, FUN, FUN.VALUE, ..., USE.NAMES = TRUE, cl = NULL)
pbreplicate(n, expr, simplify = "array", cl = NULL)

pbmapply(FUN, ..., MoreArgs = NULL, SIMPLIFY = TRUE, USE.NAMES = TRUE)
pb.mapply(FUN, dots, MoreArgs)
pbMap(f, ...)

pbtapply(X, INDEX, FUN = NULL, ..., default = NA, simplify = TRUE, cl = NULL)

pbby(data, INDICES, FUN, ..., simplify = TRUE, cl = NULL)



For pbsapply, pblapply, and pbwalk a vector (atomic or list) or an expressions vector (other objects including classed objects will be coerced by as.list.) For pbapply an array, including a matrix. For pbtapply an R object for which a split method exists. Typically vector-like, allowing subsetting with [.


A vector giving the subscripts which the function will be applied over. 1 indicates rows, 2 indicates columns, c(1,2) indicates rows and columns.

FUN, f

The function to be applied to each element of X: see apply, sapply, and lapply. In the case of functions like +, %*%, etc., the function name must be backquoted or quoted. If FUN is NULL, pbtapply returns a vector which can be used to subscript the multi-way array pbtapply normally produces.


Optional arguments to FUN and also to underlying functions (e.g. parLapply and mclapply when cl is not NULL).


List of arguments to vectorize over (vectors or lists of strictly positive length, or all of zero length); see .mapply.


Environment to be used.


A (generalized) vector; a template for the return value from FUN. See 'Details' for vapply.

simplify, SIMPLIFY

Logical; should the result be simplified to a vector or matrix if possible? pbtapply returns an array of mode "list" (in other words, a list with a dim attribute) when FALSE; if TRUE (the default), then if FUN always returns a scalar, pbtapply returns an array with the mode of the scalar.


Logical; if TRUE and if X is character, use X as names for the result unless it had names already.


Logical, indicating whether to apply the function to all values.


Number of replications.


Expression (language object, usually a call) to evaluate repeatedly.


A cluster object created by makeCluster, or an integer to indicate number of child-processes (integer values are ignored on Windows) for parallel evaluations (see Details on performance). It can also be "future" to use a future backend (see Details), NULL (default) refers to sequential evaluation.


A list of other arguments to FUN.


A list of one or more factors, each of same length as X. The elements are coerced to factors by as.factor.


A factor or a list of factors, each of length nrow(data).


An R object, normally a data frame, possibly a matrix.


Only in the case of simplification to an array, the value with which the array is initialized as array(default, dim = ..). Before R 3.4.0, this was hard coded to array()'s default NA. If it is NA (the default), the missing value of the answer type, e.g. NA_real_, is chosen (as.raw(0) for "raw"). In a numerical case, it may be set, e.g., to FUN(integer(0)), e.g., in the case of FUN = sum to 0 or 0L.


The behavior of the progress bar is controlled by the option type in pboptions, it can take values c("txt", "win", "tk", "none",) on Windows, and c("txt", "tk", "none",) on Unix systems.

Other options have elements that are arguments used in the functions timerProgressBar, txtProgressBar, and tkProgressBar. See pboptions for how to conveniently set these.

Parallel processing can be enabled through the cl argument. parLapply is called when cl is a 'cluster' object, mclapply is called when cl is an integer. Showing the progress bar increases the communication overhead between the main process and nodes / child processes compared to the parallel equivalents of the functions without the progress bar. The functions fall back to their original equivalents when the progress bar is disabled (i.e. getOption("pboptions")$type == "none" or dopb() is FALSE). This is the default when interactive() if FALSE (i.e. called from command line R script).

When doing parallel processing, other objects might need to pushed to the workers, and random numbers must be handled with care (see Examples).

Updating the progress bar with mclapply can be slightly slower compared to using a Fork cluster (i.e. calling makeForkCluster). Care must be taken to set appropriate random numbers in this case.

Note the use_lb option (see pboptions) for using load balancing when running in parallel clusters. If using mclapply, the ... passes arguments to the underlying function for further control.

pbwalk is similar to pblapply but it calls FUN only for its side-effect and returns the input X invisibly (this behavior is modeled after 'purrr::walk').

Note that when cl = "future", you might have to specify the future.seed argument (passed as part of ...) when using random numbers in parallel.

Note also that if your code prints messages or you encounter warnings during execution, the condition messages might cause the progress bar to break up and continue on a new line.


Similar to the value returned by the standard *apply functions.

A progress bar is showed as a side effect.


Progress bar can add an overhead to the computation.


Peter Solymos <solymos@ualberta.ca>

See Also

Progress bars used in the functions: txtProgressBar, tkProgressBar, timerProgressBar

Sequential *apply functions: apply, sapply, lapply, replicate, mapply, .mapply, tapply

Parallel *apply functions from package 'parallel': parLapply, mclapply.

Setting the options: pboptions

Conveniently add progress bar to for-like loops: startpb, setpb, getpb, closepb


## --- simple linear model simulation ---
n <- 200
x <- rnorm(n)
y <- rnorm(n, crossprod(t(model.matrix(~ x)), c(0, 1)), sd = 0.5)
d <- data.frame(y, x)
## model fitting and bootstrap
mod <- lm(y ~ x, d)
ndat <- model.frame(mod)
B <- 100
bid <- sapply(1:B, function(i) sample(nrow(ndat), nrow(ndat), TRUE))
fun <- function(z) {
    if (missing(z))
        z <- sample(nrow(ndat), nrow(ndat), TRUE)
    coef(lm(mod$call$formula, data=ndat[z,]))

## standard '*apply' functions
system.time(res1 <- lapply(1:B, function(i) fun(bid[,i])))
system.time(res2 <- sapply(1:B, function(i) fun(bid[,i])))
system.time(res3 <- apply(bid, 2, fun))
system.time(res4 <- replicate(B, fun()))

## 'pb*apply' functions
## try different settings:
## "none", "txt", "tk", "win", "timer"
op <- pboptions(type = "timer") # default
system.time(res1pb <- pblapply(1:B, function(i) fun(bid[,i])))

pboptions(type = "txt")
system.time(res2pb <- pbsapply(1:B, function(i) fun(bid[,i])))

pboptions(type = "txt", style = 1, char = "=")
system.time(res3pb <- pbapply(bid, 2, fun))

pboptions(type = "txt", char = ":")
system.time(res4pb <- pbreplicate(B, fun()))

## Not run: 
## parallel evaluation using the parallel package
## (n = 2000 and B = 1000 will give visible timing differences)

cl <- makeCluster(2L)
clusterExport(cl, c("fun", "mod", "ndat", "bid"))

## parallel with no progress bar: snow type cluster
## (RNG is set in the main process to define the object bid)
system.time(res1cl <- parLapply(cl = cl, 1:B, function(i) fun(bid[,i])))
system.time(res2cl <- parSapply(cl = cl, 1:B, function(i) fun(bid[,i])))
system.time(res3cl <- parApply(cl, bid, 2, fun))

## parallel with  progress bar: snow type cluster
## (RNG is set in the main process to define the object bid)
system.time(res1pbcl <- pblapply(1:B, function(i) fun(bid[,i]), cl = cl))
system.time(res2pbcl <- pbsapply(1:B, function(i) fun(bid[,i]), cl = cl))
## (RNG needs to be set when not using bid)
parallel::clusterSetRNGStream(cl, iseed = 0L)
system.time(res4pbcl <- pbreplicate(B, fun(), cl = cl))
system.time(res3pbcl <- pbapply(bid, 2, fun, cl = cl))


if (.Platform$OS.type != "windows") {
    ## parallel with no progress bar: multicore type forking
    ## (mc.set.seed = TRUE in parallel::mclapply by default)
    system.time(res2mc <- mclapply(1:B, function(i) fun(bid[,i]), mc.cores = 2L))
    ## parallel with  progress bar: multicore type forking
    ## (mc.set.seed = TRUE in parallel::mclapply by default)
    system.time(res1pbmc <- pblapply(1:B, function(i) fun(bid[,i]), cl = 2L))
    system.time(res2pbmc <- pbsapply(1:B, function(i) fun(bid[,i]), cl = 2L))
    system.time(res4pbmc <- pbreplicate(B, fun(), cl = 2L))

## End(Not run)

## --- Examples taken from standard '*apply' functions ---

## --- sapply, lapply, and replicate ---

require(stats); require(graphics)

x <- list(a = 1:10, beta = exp(-3:3), logic = c(TRUE,FALSE,FALSE,TRUE))
# compute the list mean for each list element
pblapply(x, mean)
pbwalk(x, mean)
# median and quartiles for each list element
pblapply(x, quantile, probs = 1:3/4)
pbsapply(x, quantile)
i39 <- sapply(3:9, seq) # list of vectors
pbsapply(i39, fivenum)
pbvapply(i39, fivenum,
       c(Min. = 0, "1st Qu." = 0, Median = 0, "3rd Qu." = 0, Max. = 0))

## sapply(*, "array") -- artificial example
(v <- structure(10*(5:8), names = LETTERS[1:4]))
f2 <- function(x, y) outer(rep(x, length.out = 3), y)
(a2 <- pbsapply(v, f2, y = 2*(1:5), simplify = "array"))
a.2 <- pbvapply(v, f2, outer(1:3, 1:5), y = 2*(1:5))
stopifnot(dim(a2) == c(3,5,4), all.equal(a2, a.2),
          identical(dimnames(a2), list(NULL,NULL,LETTERS[1:4])))

summary(pbreplicate(100, mean(rexp(10))))

## use of replicate() with parameters:
foo <- function(x = 1, y = 2) c(x, y)
# does not work: bar <- function(n, ...) replicate(n, foo(...))
bar <- function(n, x) pbreplicate(n, foo(x = x))
bar(5, x = 3)

## --- apply ---

## Compute row and column sums for a matrix:
x <- cbind(x1 = 3, x2 = c(4:1, 2:5))
dimnames(x)[[1]] <- letters[1:8]
pbapply(x, 2, mean, trim = .2)
col.sums <- pbapply(x, 2, sum)
row.sums <- pbapply(x, 1, sum)
rbind(cbind(x, Rtot = row.sums), Ctot = c(col.sums, sum(col.sums)))

stopifnot( pbapply(x, 2, is.vector))

## Sort the columns of a matrix
pbapply(x, 2, sort)

## keeping named dimnames
names(dimnames(x)) <- c("row", "col")
x3 <- array(x, dim = c(dim(x),3),
	    dimnames = c(dimnames(x), list(C = paste0("cop.",1:3))))
identical(x,  pbapply( x,  2,  identity))
identical(x3, pbapply(x3, 2:3, identity))

##- function with extra args:
cave <- function(x, c1, c2) c(mean(x[c1]), mean(x[c2]))
pbapply(x, 1, cave,  c1 = "x1", c2 = c("x1","x2"))

ma <- matrix(c(1:4, 1, 6:8), nrow = 2)
pbapply(ma, 1, table)  #--> a list of length 2
pbapply(ma, 1, stats::quantile) # 5 x n matrix with rownames

stopifnot(dim(ma) == dim(pbapply(ma, 1:2, sum)))

## Example with different lengths for each call
z <- array(1:24, dim = 2:4)
zseq <- pbapply(z, 1:2, function(x) seq_len(max(x)))
zseq         ## a 2 x 3 matrix
typeof(zseq) ## list
dim(zseq) ## 2 3
pbapply(z, 3, function(x) seq_len(max(x)))
# a list without a dim attribute

## --- mapply and .mapply ---

pbmapply(rep, 1:4, 4:1)
pbmapply(rep, times = 1:4, x = 4:1)
pbmapply(rep, times = 1:4, MoreArgs = list(x = 42))
pbmapply(function(x, y) seq_len(x) + y,
       c(a =  1, b = 2, c = 3),  # names from first
       c(A = 10, B = 0, C = -10))
word <- function(C, k) paste(rep.int(C, k), collapse = "")
utils::str(pbmapply(word, LETTERS[1:6], 6:1, SIMPLIFY = FALSE))

          dots = list(1:4, 4:1),
          MoreArgs = list())
          dots = list(times = 1:4, x = 4:1),
          MoreArgs = list())
          dots = list(times = 1:4),
          MoreArgs = list(x = 42))
pb.mapply(function(x, y) seq_len(x) + y,
          dots = list(c(a =  1, b = 2, c = 3),  # names from first
                      c(A = 10, B = 0, C = -10)),
          MoreArgs = list())

## --- Map ---

pbMap(`+`, 1,         1 : 3) ;         1 + 1:3

## --- eapply ---

env <- new.env(hash = FALSE)
env$a <- 1:10
env$beta <- exp(-3:3)
env$logic <- c(TRUE, FALSE, FALSE, TRUE)
pbeapply(env, mean)
unlist(pbeapply(env, mean, USE.NAMES = FALSE))
pbeapply(env, quantile, probs = 1:3/4)
pbeapply(env, quantile)

## --- tapply ---

groups <- as.factor(rbinom(32, n = 5, prob = 0.4))
pbtapply(groups, groups, length) #- is almost the same as

## contingency table from data.frame : array with named dimnames
pbtapply(warpbreaks$breaks, warpbreaks[,-1], sum)
pbtapply(warpbreaks$breaks, warpbreaks[, 3, drop = FALSE], sum)

n <- 17; fac <- factor(rep_len(1:3, n), levels = 1:5)
pbtapply(1:n, fac, sum)
pbtapply(1:n, fac, sum, default = 0) # maybe more desirable
pbtapply(1:n, fac, sum, simplify = FALSE)
pbtapply(1:n, fac, range)
pbtapply(1:n, fac, quantile)
pbtapply(1:n, fac, length) ## NA's
pbtapply(1:n, fac, length, default = 0) # == table(fac)

## example of ... argument: find quarterly means
pbtapply(presidents, cycle(presidents), mean, na.rm = TRUE)

ind <- list(c(1, 2, 2), c("A", "A", "B"))
pbtapply(1:3, ind) #-> the split vector
pbtapply(1:3, ind, sum)

## Some assertions (not held by all patch propsals):
nq <- names(quantile(1:5))
  identical(pbtapply(1:3, ind), c(1L, 2L, 4L)),
  identical(pbtapply(1:3, ind, sum),
            matrix(c(1L, 2L, NA, 3L), 2, dimnames = list(c("1", "2"), c("A", "B")))),
  identical(pbtapply(1:n, fac, quantile)[-1],
            array(list(`2` = structure(c(2, 5.75, 9.5, 13.25, 17), .Names = nq),
                 `3` = structure(c(3, 6, 9, 12, 15), .Names = nq),
                 `4` = NULL, `5` = NULL), dim=4, dimnames=list(as.character(2:5)))))

## --- by ---

pbby(warpbreaks[, 1:2], warpbreaks[,"tension"], summary)
pbby(warpbreaks[, 1],   warpbreaks[, -1],       summary)
pbby(warpbreaks, warpbreaks[,"tension"],
   function(x) lm(breaks ~ wool, data = x))
tmp <- with(warpbreaks,
            pbby(warpbreaks, tension,
               function(x) lm(breaks ~ wool, data = x)))
sapply(tmp, coef)

pbapply documentation built on July 9, 2023, 7:41 p.m.

Related to pbapply in pbapply...