CODEWITHSUNDEEP
r
Adamaki
Adamaki

Reputation: 47

Summing column according to consecutive factor instances

I have a dataset of fish tag signals, and I want to calculate the duration of different behaviours based on swimming speed, e.g. static, cruise, burst, so I can calculate behaviour state frequencies. I have done this using a for loop, but it's very slow on my large datasets. I'm sure this can be done using one of R's apply functions, but I can't figure out how to do it.

This is what my data looks like:

    Period PEN SEC BLSEC     BS  BScount CountTF  BSdur
380   7045   7   7 0.204 cruise        2   FALSE     NA
381   7045   7   7 0.694 cruise        3   FALSE     NA
382   7045   7   7 0.325 cruise        4    TRUE     21
383   7045   7   7 0.000 static        1    TRUE      7
384   7045   7   7 0.197 cruise        1   FALSE     NA
385   7045   7   7 0.312 cruise        2   FALSE     NA
386   7045   7   7 0.242 cruise        3    TRUE     21
387   7045   7   7 0.096 static        1    TRUE      7
388   7045   7   7 0.274 cruise        1   FALSE     NA
389   7045   7   7 0.268 cruise        2   FALSE     NA
390   7045   7   7 0.312 cruise        3   FALSE     NA
391   7045   7   7 0.694 cruise        4   FALSE     NA
392   7045   7   7 0.268 cruise        5   FALSE     NA

SEC is the number of seconds between tag pings (it's not always 7!), BLSEC is body lengths per second (i.e. normalised distance swam by the fish between tag pings). I've calculated BS, BScount and CountTF by doing:

static = 0.1
cruise = 1

bsffile$BS <- ifelse(bsffile$BLSEC <= static, 'static', ifelse(bsffile$BLSEC > static & bsffile$BLSEC <= cruise, 'cruise', 'burst'))
bsffile$BScount <- sequence(rle(bsffile$BS)$lengths)
bsffile$CountTF <- c(ifelse(diff(bsffile$BScount, 1, 1) < 1, T, F), F)

BSdur is the sum of SECs for consecutive behaviour states. I calculated it using:

bssum <- 0

for (i in 1:nrow(bsffile)){
  bssum <- bssum + bsffile[i, 'SEC']
  if(bsffile[i, 'CountTF'] == T & is.na(bsffile[i, 'SEC']) == F){
    bsffile[i,'BSdur'] <- bssum
    bssum <- 0
  } else {
    bsffile[i,'BSdur'] <- NA    
  }
}

It takes about five minutes to run on my dataset. Any suggestions how I can make this faster, e.g using one of the apply functions?

Here is some dput to play with:

structure(list(Period = c(7045, 7045, 7045, 7045, 7045, 7045, 
7045, 7045, 7045, 7045, 7045, 7045, 7045, 7045, 7045, 7045, 7045, 
7045, 7045, 7045, 7045), PEN = structure(c(1L, 1L, 1L, 1L, 1L, 
1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L
), .Label = c("7", "8"), class = "factor"), SEC = c(7, 7, 7, 
7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 8, 7, 7, 7, 7, 7), BLSEC = c(0.204, 
0.694, 0.325, 0, 0.197, 0.312, 0.242, 0.096, 0.274, 0.268, 0.312, 
0.694, 0.268, 0.541, 0.796, 0.306, 0.089, 0.93, 0.389, 0.452, 
0.917), BS = c("cruise", "cruise", "cruise", "static", "cruise", 
"cruise", "cruise", "static", "cruise", "cruise", "cruise", "cruise", 
"cruise", "cruise", "cruise", "cruise", "static", "cruise", "cruise", 
"cruise", "cruise"), BScount = c(2L, 3L, 4L, 1L, 1L, 2L, 3L, 
1L, 1L, 2L, 3L, 4L, 5L, 6L, 7L, 8L, 1L, 1L, 2L, 3L, 4L), CountTF = c(FALSE, 
FALSE, TRUE, TRUE, FALSE, FALSE, TRUE, TRUE, FALSE, FALSE, FALSE, 
FALSE, FALSE, FALSE, FALSE, TRUE, TRUE, FALSE, FALSE, FALSE, 
TRUE), BSdur = c(NA, NA, 21, 7, NA, NA, 21, 7, NA, NA, NA, NA, 
NA, NA, NA, 57, 7, NA, NA, NA, 28)), row.names = 380:400, .Names = c("Period", 
"PEN", "SEC", "BLSEC", "BS", "BScount", "CountTF", "BSdur"
), class = "data.frame")

Upvotes: 0

Views: 39

Answers (2)

akrun
akrun

Reputation: 887153

We can do this with ave from base R

df1$BSdur <- with(df1, ave(SEC, cumsum(c(TRUE, BS[-1]!= BS[-nrow(df1)])), FUN = sum)*CountTF)
df1$BSdur
#[1]  0  0 21  7  0  0 21  7  0  0  0  0  0  0  0 57  7  0  0  0 28

Upvotes: 0

Erdem Akkas
Erdem Akkas

Reputation: 2070

Easy with data.table

library(data.table)
setDT(bsffile)
bsffile[,BSdur:=ifelse(CountTF==T,sum(SEC),0),by=.(rleid(BS))]

Upvotes: 2

Related Questions