Using rle() according to other variable group R

Question

This is really a follow on from another question I posted on here a few weeks ago and got an answer for.

In my initial question I wanted to find the number of days between runoff events in a dataset. As shown in the data sample below:

Date        Runoff   No_Days
01/01/1980  0        4
02/01/1980  0        3
03/01/1980  0        2
04/01/1980  0        1
05/01/1980  4.5      0
06/01/1980  2        0
07/01/1980  0        6
08/01/1980  0        5
09/01/1980  0        4
10/01/1980  0        3
11/01/1980  0        2
12/01/1980  0        1  
13/01/1980  1.2      0
14/01/1980  0        4      
15/01/1980  0        3
16/01/1980  0        2
17/01/1980  0        1
18/01/1980  0.8      0

I managed to get to this using the following code:

DF$No_Days <-unlist(lapply(rle(DF$Runoff>0.05)$lengths,function(x) rev(seq(x:1))))
DF$No_Days <-ifelse(DF$Runoff>0.05,0,DF$No_Days)

This all works well for a single dataset i.e. one time series for one group. What I am struggling with now however is how to manipulate the above code to do the same things for a number of time series datasets in the same data.table according to a grouping variable (Soil) to get for example:

Date        Runoff   No_Days  Soil
01/01/1980  0        4        Clay
02/01/1980  0        3        Clay    
03/01/1980  0        2        Clay    
04/01/1980  0        1        Clay    
05/01/1980  4.5      0        Clay
06/01/1980  2        0        Clay
07/01/1980  0        6        Clay
08/01/1980  0        5        Clay
09/01/1980  0        4        Clay
10/01/1980  0        3        Clay
11/01/1980  0        2        Clay
12/01/1980  0        1        Clay
13/01/1980  1.2      0        Clay
14/01/1980  0        4        Clay  
15/01/1980  0        3        Clay
16/01/1980  0        2        Clay
17/01/1980  0        1        Clay
18/01/1980  0.8      0        Clay
01/01/1980  0        5        Sand
02/01/1980  0        4        Sand
03/01/1980  0        3        Sand
04/01/1980  0        2        Sand
05/01/1980  0        1        Sand
06/01/1980  2        0        Sand
07/01/1980  0        11       Sand
08/01/1980  0        10       Sand
09/01/1980  0        9        Sand
10/01/1980  0        8        Sand
11/01/1980  0        7        Sand
12/01/1980  0        6        Sand
13/01/1980  0        5        Sand
14/01/1980  0        4        Sand    
15/01/1980  0        3        Sand
16/01/1980  0        2        Sand
17/01/1980  0        1        Sand
18/01/1980  0.8      0        Sand

Currently if I run the code it does not distinguish between the different soil types and therefore does not 'restart' the sequencing after each time series.

From reading around it seems that I may need to replace lapply() in the original code to by(). I think this will work as long as rle() is first of all grouped according to Soil but I can't find any way of doing this.

So any help appreciated please!

Answer 1

If you use the data.table package, this is very easy:

install.packages("data.table")
library(data.table)
DF = data.table(DF)

DF[,No_Days:=unlist(lapply(rle(Runoff>0.05)$lengths,function(x) rev(seq(x:1)))),by=Soil]
DF[Runoff <= 0.05, No_Days:=0]

Answer 2

If you were interested in doing this in base R as well, you can use ave to get the same result. For convenience, i'll define

countdown <- function(events) {
    unlist(with(rle(events), 
        Map(function(v,l) {
            if(v) rep.int(0,l)
            else l:1}
        , values, lengths)
    ))
}

Then you would find the answer without soil type with

DF <- transform(DF, No_Days=countdown(Runoff>0.05))

and then to do grouping by soil type you could to

DF <- transform(DF, No_Days=ave(Runoff>0.05, Soil, FUN=countdown))