Original Authors: Valerie Obenchain, Martin Morgan
Presenting Author: Martin Morgan (
Date: 15 July, 2016

Efficient R code

The goal of this section is to highlight practices for writing correct, robust and efficient R code.


  1. Correct: consistent with hand-worked examples (identical(), all.equal())
  2. Robust: supports realistic inputs, e.g., 0-length vectors, NA values, …
  3. Simple: easy to understand next month; easy to describe what it does to a colleague; easy to spot logical errors; easy to enhance.
  4. Fast, or at least reasonable given the speed of modern computers.


  1. Profile
  1. Vectorize – operate on vectors, rather than explicit loops

    x <- 1:10
    log(x)     ## NOT for (i in seq_along) x[i] <- log(x[i])
    ##  [1] 0.0000000 0.6931472 1.0986123 1.3862944 1.6094379 1.7917595 1.9459101
    ##  [8] 2.0794415 2.1972246 2.3025851
  2. Pre-allocate memory, then fill in the result

    result <- numeric(10)
    result[1] <- runif(1)
    for (i in 2:length(result))
           result[i] <- runif(1) * result[i - 1]
    ##  [1] 0.9064329518 0.1066930705 0.0861235086 0.0767101194 0.0599125034
    ##  [6] 0.0131052207 0.0060250919 0.0045694322 0.0008398694 0.0006845054
  3. Hoist common sub-expressions outside of repeated calculations, so that the sub-expression is only calculated once
  1. Re-use existing, tested code

Back to top

Case Study: Pre-allocate and vectorize

Here’s an obviously inefficient function:

f0 <- function(n, a=2) {
    ## stopifnot(is.integer(n) && (length(n) == 1) &&
    ##           ! && (n > 0))
    result <- numeric()
    for (i in seq_len(n))
        result[[i]] <- a * log(i)

Use system.time() to investigate how this algorithm scales with n, focusing on elapsed time.

##    user  system elapsed 
##   0.088   0.004   0.091
n <- 1000 * seq(1, 20, 2)
t <- sapply(n, function(i) system.time(f0(i))[[3]])
plot(t ~ n, type="b")

Remember the current ‘correct’ value, and an approximate time

n <- 10000
system.time(expected <- f0(n))
##    user  system elapsed 
##   0.116   0.000   0.114
## [1] 0.000000 1.386294 2.197225 2.772589 3.218876 3.583519

Revise the function to hoist the common multiplier, a, out of the loop. Make sure the result of the ‘optimization’ and the original calculation are the same. Use the microbenchmark package to compare the two versions

f1 <- function(n, a=2) {
    result <- numeric()
    for (i in seq_len(n))
        result[[i]] <- log(i)
    a * result
identical(expected, f1(n))
## [1] TRUE
microbenchmark(f0(n), f1(n), times=5)
## Unit: milliseconds
##   expr      min        lq     mean   median       uq      max neval
##  f0(n) 93.75949 125.43283 120.7678 126.8942 127.3390 130.4134     5
##  f1(n) 97.80559  99.46675 116.5440 126.8709 129.2406 129.3359     5

Adopt a ‘pre-allocate and fill’ strategy

f2 <- function(n, a=2) {
    result <- numeric(n)
    for (i in seq_len(n))
        result[[i]] <- log(i)
    a * result
identical(expected, f2(n))
## [1] TRUE
microbenchmark(f0(n), f2(n), times=5)
## Unit: milliseconds
##   expr       min         lq       mean    median         uq      max neval
##  f0(n) 99.133356 127.296109 123.179391 128.47307 129.965368 131.0291     5
##  f2(n)  6.887973   6.981819   7.291664   7.08013   7.471796   8.0366     5

Use an *apply() function to avoid having to explicitly pre-allocate, and make opportunities for vectorization more apparent.

f3 <- function(n, a=2)
    a * sapply(seq_len(n), log)

identical(expected, f3(n))
## [1] TRUE
microbenchmark(f0(n), f2(n), f3(n), times=10)
## Unit: milliseconds
##   expr       min         lq       mean     median         uq        max
##  f0(n) 99.826475 102.051590 123.345929 131.463728 133.418350 136.598269
##  f2(n)  6.911975   7.036002   7.523884   7.272184   8.193247   8.391159
##  f3(n)  3.120228   3.167543   3.436611   3.281130   3.726726   4.354170
##  neval
##     10
##     10
##     10

Now that the code is presented in a single line, it is apparent that it could be easily vectorized. Seize the opportunity to vectorize it:

f4 <- function(n, a=2)
    a * log(seq_len(n))
identical(expected, f4(n))
## [1] TRUE
microbenchmark(f0(n), f3(n), f4(n), times=10)
## Unit: microseconds
##   expr       min         lq        mean     median         uq        max
##  f0(n) 98247.257 100166.209 120943.9480 130044.098 130644.027 131510.762
##  f3(n)  2848.699   3077.010   3101.3840   3093.171   3190.727   3231.083
##  f4(n)   204.107    206.843    294.2586    211.896    213.417   1034.029
##  neval
##     10
##     10
##     10

f4() definitely seems to be the winner. How does it scale with n? (Repeat several times)

n <- 10 ^ (5:8)                         # 100x larger than f0
t <- sapply(n, function(i) system.time(f4(i))[[3]])
plot(t ~ n, log="xy", type="b")

Any explanations for the different pattern of response?

Lessons learned:

  1. Vectorizing offers a huge improvement over iteration
  2. Pre-allocate-and-fill is very helpful when explicit iteration is required.
  3. *apply() functions help avoid need for explicit pre-allocation and make opportunities for vectorization more apparent. This may come at a small performance cost, but is generally worth it
  4. Hoisting common sub-expressions can be helpful for improving performance when explicit iteration is required.

Back to top

Iteration and restriction to manage memory

When data are too large to fit in memory, we can iterate through files in chunks or subset the data by fields or genomic positions.

Iteration - Chunk-wise - open(), read chunk(s), close(). - e.g., yieldSize argument to Rsamtools::BamFile() - Framework: GenomicFiles::reduceByYield()

Restriction - Limit to columns and / or rows of interest - Exploit domain-specific formats – BAM files and Rsamtools::ScanBamParam() – BAM files and Rsamtools::PileupParam() – VCF files and VariantAnnotation::ScanVcfParam() - Use a data base

Exercise: Counting overlaps

Iterate through files: `GenomicFiles::reduceByYield()

  1. yield a chunk
  2. map from the input chunk to a possibly transformed representation
  3. reduce mapped chunks

yield <-     # how to input the next chunk of data
    function(X, ...)

map <-       # what to do to each chunk
    function(VALUE, ..., roi)
    olaps <- findOverlaps(VALUE, roi, type="within", ignore.strand=TRUE)
    count <- tabulate(subjectHits(olaps), subjectLength(olaps))
    setNames(count, names(roi))

reduce <- `+`   # how to combine mapped chunks

Improvement: “yield factory” to keep track of how many records input

yieldFactory <-  # return a function with local state 
    n_records <- 0L
    function(X, ...) {
        aln <- readGAlignments(X)
        n_records <<- n_records + length(aln)

Regions of interest, named like the chromosomes in the bam file.

exByTx <- exonsBy(TxDb.Hsapiens.UCSC.hg19.knownGene, "tx")

fl <- "/home/ubuntu/data/vobencha/LargeData/srarchive/"
map0 <- read.delim(fl, header=FALSE, stringsAsFactors=FALSE)
seqlevels(exByTx, force=TRUE) <- setNames(map0$V1, map0$V2)

A function to iterate through a bam file.

count1 <- function(filename, roi) {
    ## Create and open BAM file
    bf <- BamFile(filename, yieldSize=1000000)
    reduceByYield(bf, yieldFactory(), map, reduce, roi=roi)

In action

bam <- "/home/ubuntu/data/vobencha/LargeData/srarchive/SRR1039508_sorted.bam"
count <- count1(bam, exByTx)

Back to top

File management

File classes

Type Example use Name Package
.bed Range annotations BedFile() rtracklayer
.wig Coverage WigFile(), BigWigFile() rtracklayer
.gtf Transcript models GTFFile() rtracklayer
makeTxDbFromGFF() GenomicFeatures
.2bit Genomic Sequence TwoBitFile() rtracklayer
.fastq Reads & qualities FastqFile() ShortRead
.bam Aligned reads BamFile() Rsamtools
.tbx Indexed tab-delimited TabixFile() Rsamtools
.vcf Variant calls VcfFile() VariantAnnotation
## rtracklayer menagerie
##  [1] "UCSCFile"         "GFFFile"          "BEDFile"         
##  [4] "WIGFile"          "BigWigFile"       "ChainFile"       
##  [7] "TwoBitFile"       "FastaFile"        "TabSeparatedFile"
## [10] "CompressedFile"   "GFF1File"         "GFF2File"        
## [13] "GFF3File"         "BEDGraphFile"     "BED15File"       
## [16] "BEDPEFile"        "BWFile"           "2BitFile"        
## [19] "GTFFile"          "GVFFile"          "GZFile"          
## [22] "BGZFile"          "BZ2File"          "XZFile"


Managing a collection of files

*FileList() classes



Back to top

Parallel evaluation

A couple of caveats -

Iteration / restriction techniques keep the memory requirements under control while parallel evaluation distributes computational load across nodes. Keep in mind that parallel computations are still restricted by the amount of memory available on each node.

There is overhead in setting up and tearing down a cluster and more so when computing in distributed memory. For small calculations, the parallel overhead may outweigh the benefits with no improvement in performance.

Jobs that benefit the most from parallel execution are CPU-intensive and operate on data chunks that fits into memory.


BiocParallel provides a standardized interface for parallel evaluation and supports the major parallel computing styles: forks and processes on a single computer, ad hoc clusters, batch schedulers and cloud computing. By default, BiocParallel chooses a parallel back-end appropriate for the OS and is supported across Unix, Mac and Windows.

General ideas:

Exercise: Sleeping serially and in parallel

This small example motivates the use of parallel execution and demonstrates how bplapply() can be a drop in for lapply.

Use system.time() to explore how long this takes to execute as n increases from 1 to 10. Use identical() and microbenchmark to compare alternatives f0() and f1() for both correctness and performance.

fun sleeps for 1 second, then returns i.


fun <- function(i) {

## serial
f0 <- function(n)
    lapply(seq_len(n), fun)

## parallel
f1 <- function(n)
    bplapply(seq_len(n), fun)

Back to top

Exercise: error handling and BPREDO

BiocParallel “catches and returns” errors along with successful results. This exercise demonstrates how to access the traceback() of a failed task and how to re-run the failed tasks with ‘BPREDO’. Full details on error handling, logging and debugging are in the Errors, Logs and Debugging vignette.

param <- MulticoreParam(workers = 3)

Call the sqrt() function across ‘X’; the second element is a character and will throw and error.

X <- list(1, "2", 3)
res <- bplapply(X, sqrt, BPPARAM = param)
## Error: BiocParallel errors
##   element index: 2
##   first error: non-numeric argument to mathematical function

It’s also possible to catch the error and partially evaluated result

res <- bptry(bplapply(X, sqrt, BPPARAM=param))
## [[1]]
## [1] 1
## [[2]]
## <remote_error in FUN(...): non-numeric argument to mathematical function>
## traceback() available as 'attr(x, "traceback")'
## [[3]]
## [1] 1.732051

Re-run the failed results by repeating the call to bplapply() this time with corrected input data and the partial results as ‘BPREDO’. Only the failed values are re-run.

X.redo <- list(1, 2, 3)
bplapply(X.redo, sqrt, BPREDO = res)
## resuming previous calculation ...
## [[1]]
## [1] 1
## [[2]]
## [1] 1.414214
## [[3]]
## [1] 1.732051

Alternatively, switch to a SerialParam() and debug the specific element that caused the error.

> fun = function(i) { browser(); sqrt(i) }
> bplapply(X, fun, BPREDO=res, BPPARAM=SerialParam())
resuming previous calculation ... 
Called from: FUN(...)
debug at #1: sqrt(i)
Browse[2]> i
[1] "2"
Browse[2]> i = 2
Browse[2]> c
[1] 1

[1] 1.414214

[1] 1.732051

Back to top

Exercise: logging

BiocParallel uses the futile.logger package for logging. The package has a flexible system for filtering messages of varying severity thresholds such as INFO, DEBUG, ERROR etc. (For a list of all thresholds see the ?bpthreshold man page.) BiocParallel captures messages written in futile.logger format as well as messages written to stdout and stderr.

This function does some argument checking and has DEBUG, WARN and INFO-level log messages.

FUN <- function(i) {
  flog.debug(paste0("value of 'i': ", i))

  if (!length(i)) {
      flog.warn("'i' is missing")
  } else if (!is(i, "numeric")) {"coercing to numeric")
  } else {

Turn logging on in the param and set the threshold to WARN.

param <- SnowParam(3, log = TRUE, threshold = "WARN")
bplapply(list(1, "2", integer()), FUN, BPPARAM = param)

Lower the threshold to INFO and DEBUG (i.e., use bpthreshold<-) to see how messages are filtered on severity.

Back to top

Exercise: Worker timeout

For long running jobs or untested code it can be useful to set a time limit. The timeout field is the time, in seconds, allowed for each worker to complete a task. If a task takes longer than timeout the worker returns an error.

timeout can be set during param construction,

param <- SnowParam(timeout = 20)
## class: SnowParam
##   bpisup: FALSE; bpworkers: 6; bptasks: 0; bpjobname: BPJOB
##   bplog: FALSE; bpthreshold: INFO; bpstopOnError: TRUE
##   bptimeout: 20; bpprogressbar: FALSE
##   bpRNGseed: 
##   bplogdir: NA
##   bpresultdir: NA
##   cluster type: SOCK

or with the setter:

bptimeout(param) <- 2 
## class: SnowParam
##   bpisup: FALSE; bpworkers: 6; bptasks: 0; bpjobname: BPJOB
##   bplog: FALSE; bpthreshold: INFO; bpstopOnError: TRUE
##   bptimeout: 2; bpprogressbar: FALSE
##   bpRNGseed: 
##   bplogdir: NA
##   bpresultdir: NA
##   cluster type: SOCK

Use this function to explore different _timeout_s over a numeric vector of ‘X’ values with bplapply(). ‘X’ values less than timeout return successfully while those over threshold return an error.

fun <- function(i) {

Back to top

Exercise: Counting overlaps in parallel

Distribute files over workers: GenomicFiles::reduceByFile()

The previous counting example used GenomicFiles::reduceByYield() which operates on a single file and implements a yield, map, reduce paradigm. In this exercise we’ll use GenomicFiles::reduceByFile() which uses bplaply() under the hood to operate on multiple files in parallel.

Primary arguments to reduceByFile() are a set of files and a set of ranges. Files are sent to the workers and data subsets are extracted based on the ranges. The bulk of the work is done in the MAP function and an optional REDUCE function combines the output on each worker.


On Unix or Mac, configure a MulticoreParam() with 4 workers. Turn on logging and set a timeout of 60 seconds.

param <- MulticoreParam(4, log = TRUE, timeout = 60)

On Windows do the same with SnowParam():

param <- SnowParam(4, log = TRUE, timeout = 60)

Point to the collection of bam files.

fls <- dir("/home/ubuntu/data/vobencha/LargeData/copynumber", ".bam$", full=TRUE)
names(fls) <- basename(fls)
bfl <- BamFileList(fls)

Defining ranges (region of interest) restricts the amount of data on the workers and keeps memory requirements under control. We’ll use a set of ranges on the Major Histocompatibility Complex locus on chromosome 6.

ranges <- GRanges("chr6", IRanges(c(28477797, 29527797, 32448354),
                                  c(29477797, 30527797, 33448354)))

The MAP function reads in records and counts overlaps. readGAlignments() reads in bam records that overlap with any portion of the ranges defined in the scanBamParam (i.e., they could be overlapping the start or the end). Once we’ve got the records in R, we want to count only those that fall ‘within’ the ranges.

MAP <- function(range, file, ...) {
    library(GenomicAlignments)         ## readGAlignments(), ScanBamParam()
    param = ScanBamParam(which=range)  ## restriction
    gal = readGAlignments(file, param=param)
    ## log messages"file: ", basename(file)))
    flog.debug(paste0("records: ", length(gal)))
    ## overlaps
    olaps <- findOverlaps(gal, range, type="within", ignore.strand=TRUE)
    tabulate(subjectHits(olaps), subjectLength(olaps))

Count …

cts <- reduceByFile(ranges, fls, MAP, BPPARAM = param)

The result is a list the same length as the number of files.


Each list element is the length of the number of ranges.


Tables of counts for each range are extracted with ‘[[’:


Back to top

Other resources

Back to top


Back to top