Vectorization / acceleration of code with nested loops

Question

The code below generates the desired result. However, the lack of vectorization means that it works very slowly. How can I speed it up?

I have cited the results of dput from some of the indicative data.

dput s dput

• StandRef input

   structure(list(id = structure(c(43L, 50L, 17L, 45L, 9L, 5L, 49L, 33L, 48L, 39L, 71L, 64L, 44L, 47L, 58L, 24L, 15L, 37L, 14L, 11L, 26L, 57L, 4L, 30L, 72L, 21L, 23L, 60L, 38L, 59L, 29L, 19L, 6L, 46L, 36L, 3L, 63L, 55L, 51L, 35L, 10L, 7L, 16L, 73L, 42L, 52L, 41L, 27L, 25L, 61L, 20L, 70L, 53L, 18L, 31L, 22L, 1L, 8L, 2L, 40L, 65L, 67L, 28L, 56L, 13L, 32L, 54L, 66L, 68L, 34L, 12L, 69L, 62L), .Label = c("ID 1009445", "ID 120763", "ID 133883", "ID 136398", "ID 171850", "ID 192595", "ID 197597", "ID 216406", "ID 21888", "ID 230940", "ID 23777", "ID 282791", "ID 306348", "ID 309745", "ID 326928", "ID 344897", "ID 34974", "ID 350157", "ID 391831", "ID 402479", "ID 43010", "ID 484078", "ID 484697", "ID 537134", "ID 562259", "ID 562455", "ID 567042", "ID 572866", "ID 578945", "ID 595683", "ID 59759", "ID 598460", "ID 603611", "ID 603757", "ID 607991", "ID 60976", "ID 622720", "ID 646989", "ID 656144", "ID 668807", "ID 669435", "ID 720522", "ID 740555", "ID 745499", "ID 746001", "ID 783969", "ID 78979", "ID 792426", "ID 793541", "ID 797860", "ID 806559", "ID 810517", "ID 826054", "ID 837609", "ID 839287", "ID 867918", "ID 869788", "ID 875380", "ID 876870", "ID 882220", "ID 893116", "ID 895909", "ID 899050", "ID 900143", "ID 908100", "ID 912185", "ID 916371", "ID 916620", "ID 957879", "ID 966195", "ID 993247", "ID 998911", "ID 999610"), class = "factor"), region = c(1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L), location = c(259090L, 559306L, 2227063L, 2369217L, 4026978L, 4211264L, 4679449L, 5105226L, 5106345L, 5344670L, 5473601L, 5476528L, 5871970L, 6461228L, 6700029L, 6708265L, 7639959L, 9297695L, 10254788L, 10328812L, 11102816L, 11568295L, 11720437L, 12843457L, 14012506L, 14156669L, 14632300L, 14641938L, 15298211L, 15468425L, 15534406L, 16279682L, 16699353L, 17226952L, 17320785L, 269017L, 453097L, 828833L, 954610L, 954842L, 1066378L, 1217332L, 1253530L, 1277716L, 1292857L, 1337952L, 1439657L, 1452989L, 1712345L, 1758035L, 2601630L, 2640359L, 2778095L, 3151129L, 3369931L, 3399080L, 3529525L, 3810217L, 3821120L, 3841588L, 3901557L, 4111633L, 4220440L, 4528632L, 4665450L, 5099307L, 5260242L, 5958770L, 5966356L, 6137405L, 6246065L, 6297231L, 6807949L)), .Names = c("id", "region", "location"), class = "data.frame", row.names = c(NA, -73L)) 

• Two sample inputs

Example 1

  structure(list(region = c(1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 2L), begin = c(0L, 2259252L, 5092077L, 9158205L, 0L, 135094L, 941813L, 5901391L, 6061324L), finish = c(2259252L, 5092077L, 9158205L, 20463033L, 135094L, 941813L, 5901391L, 6061324L, 7092402L), sed = c(3.98106154985726, 7.51649828394875, 5.15440228627995, 2.67456624889746, 7.54309412557632, 4.17413910385221, 7.47043058509007, 6.13362524658442, 1.00084994221106)), .Names = c("region", "begin", "finish", "sed"), class = "data.frame", row.names = c(NA, -9L)) 

Sample 2

  structure(list(region = c(1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 2L), begin = c(0L, 2253252L, 7091077L, 9120205L, 0L, 135094L, 941813L, 5901391L, 6061324L), finish = c(2253252L, 7091077L, 9120205L, 17463033L, 135094L, 941813L, 5901391L, 6061324L, 7092402L), sed = c(3.31830840984048, 1.38014704208403, 6.13049140975458, 2.10349875097134, 0.48170587509345, 0.13058713509175, 9.13509713513509, 6.13047153058701, 3.81734081501503)), .Names = c("region", "begin", "finish", "sed"), class = "data.frame", row.names = c(NA, -9L)) 

Unclassified code

 matchLocationsToRegions <- function(path) { # get list of data files (around 500 of these; only dput of 2 given: sample262519 and sample252519) setwd(path,sep="",collapse=NULL) data_files <- list.files() # read in template file with complete regional boundaries standRef <- read.table(paste(path, "StandRef.txt",sep="",collapse=NULL), header=TRUE, sep="\t") # pre-allocate a df with row dimensions of standRef and num of columns according to num of data files sediment.df <- as.data.frame(matrix(NA,nrow=nrow(standRef),ncol=length(data_files))) colnames(sediment.df) <- data_files rownames(sediment.df) <- standRef[,1] # create a counter for columns filled col_counter <- 1 for (file in data_files) { # read in current, processed data sample <- read.table(file, header=TRUE, sep="\t") # pre-allocate vectors for sedimentation data vector sed <- rep(NA, nrow(standRef)) # create a variable to track end boundary for a particular sample_ID end_tracker <- 1 index <- unlist(lapply (unique(standRef$region), function(reg) { reg.filter <- which(standRef$region == reg) samp.filter <- which(sample$region == reg) samp.filter[cut(standRef$location[reg.filter],c(0L,sample$finish[samp.filter]),labels=F)] })) sed <- sample$sed[index] # fill in next, unfilled column of relevant df with data from relevant vector sediment.df[col_counter] <- sed # update column counter variable col_counter <- col_counter + 1 } # save df as a table write.table(sediment.df,file="samples_sed.txt", row.names=TRUE, sep="\t") } 

Running Rprof showed that "scan" "read.table" "matchLocationsToRegions" and "type.convert" "read.table" "matchLocationsToRegions" predominant runtime. Presumably, there is a bottleneck for looping along this line:

 sample <- read.table(file, header=TRUE, sep="\t") 

Update: the for loop by region has been replaced with much faster executable code (h / t Simon Urbanek). However, the rest is pretty slow.

Answer 1

You can easily remove the loop:

 sediment.df <- as.data.frame(lapply(data_files, function(file) { sample <- read.table(file, header=TRUE, sep="\t") index <- unlist(lapply (unique(standRef$region), function(reg) { reg.filter <- which(standRef$region == reg) samp.filter <- which(sample$region == reg) samp.filter[cut(standRef$location[reg.filter],c(0L,sample$finish[samp.filter]),labels=F)] })) sample$sed[index] })) colnames(sediment.df) <- data_files rownames(sediment.df) <- standRef[,1] 

However, it is unlikely that a lot of time is spent on read.table , so you can consider a) using scan , b) creating just one file with all the selections (for example, use an additional column to determine the selection), so you don't need to upload a lot of files.