extracting first and last positions in a dataset

Question

I have this dataset that I'm trying to transform to get the "from" and "to" positions within a particular grouping of data points that pass a test.

Here's how the data looks:

pos <- seq(from = 10, to = 100, by = 10)
test <- c(1, 1, 1, 0, 0, 0, 1, 1, 1, 0)
df <- data.frame(pos, test)

So you can see that positions 10, 20, and 30, as well as 70, 80, and 90 pass the test (b/c test = 1) but the rest of the points don't. The answer I'm looking for would be a data frame that looks something like the "answer" data frame in the code below:

peaknum <- c(1, 2)
from <- c(10, 70)
to <- c(30, 90)
answer <- data.frame(peaknum, from, to)

Any suggestions as to how I can transform the dataset? I'm stumped.

Thanks, Steve

Answer 1

We can use data.table. Use the rleid function to create the run-length group ids ('peaknum') based on the adjacent values that are same 'test'. Using 'peaknum' as grouping variable, we get the 'min' and 'max' of 'pos', while specifying the 'i' as 'test==1' to subset the rows. If needed, the 'peaknum' values can be changed to the sequence ('seq_len(.N)`).

library(data.table)
setDT(df)[, peaknum:= rleid(test)][test==1, 
   list(from=min(pos), to=max(pos)) ,peaknum][, peaknum:= seq_len(.N)]
#   peaknum from to
#1:       1   10 30
#2:       2   70 90

Answer 2

We can do it with dplyr, though separating the nodes is a little ugly:

library(dplyr)
df %>% group_by(peaknum = rep(seq(rle(test)[['lengths']]), rle(test)[['lengths']])) %>% 
  filter(test == 1) %>% 
  summarise(from = min(pos), 
            to = max(pos)) %>%
  mutate(peaknum = seq_along(peaknum))

# Source: local data frame [2 x 3]

#   peaknum  from    to
#     (int) (dbl) (dbl)
# 1       1    10    30
# 2       2    70    90

What it does:

• the first group_by uses rle to add a column that is a sequence along the repeated numbers in test, and groups it for summarise later;
• filter chops rows down to only those where test is 1
• summarise collapses the groups and adds max and min for each,
• and lastly mutate cleans up the numbering of peaknum.