Maximums of multiple data.table subsets - r

Question

Given is a data.table with base data, startIndex of the subsets, duration of subsets. For each subset each duration is applied.
base <- data.table(idx=c(1,2,3,4,5,6,7,8,9,10), val=c(11,12,13,14,15,16,17,18,19,20))
startIndex <- c(2, 4, 7, 9)
duration <- c(1,2,3)
Is there some elegant way to get the maximum val per subset defined by startIndex and duration, with the result shown below? For example, the first subset is defined by startIndex=2 and duration=1, which means that the max between index 2 and 3 is 13.
Result:
idxStart idxEnd max
1: 2 3 13
2: 4 5 15
3: 7 8 18
4: 2 4 14
5: 4 6 16
6: 7 9 19
7: 2 5 15
8: 4 7 17
9: 7 10 20
thx a lot.

Answer 1

Here is a data.table approach using a non-equi join. First, use expand.grid for combinations of start index and duration. Then, calculate the end index for each row. Then join with your base, where the index idx falls between the start and end, and keep the maximum val.
library(data.table)
dt <- data.table(expand.grid(idxStart = startIndex, Duration = duration))
dt[ , idxEnd := idxStart + Duration][
base, Max := max(val), on = .(idxStart <= idx, idxEnd >= idx), by = .EACHI]
Output
idxStart Duration idxEnd Max
1: 2 1 3 13
2: 4 1 5 15
3: 7 1 8 18
4: 9 1 10 20
5: 2 2 4 14
6: 4 2 6 16
7: 7 2 9 19
8: 9 2 11 20
9: 2 3 5 15
10: 4 3 7 17
11: 7 3 10 20
12: 9 3 12 20

Answer 2

I can't think of a particularly elegant solution here, but I think a map function should get the job done. This is brute forcing each combination through so there may be a more efficient solution, but it should work.
library(data.table)
base <- data.table(idx=c(1,2,3,4,5,6,7,8,9,10), val=c(11,12,13,14,15,16,17,18,19,20))
startIndex <- c(2, 4, 7, 9)
duration <- c(1,2,3)
combos <- expand.grid(startIndex = startIndex,
duration = duration) %>%
mutate(endIndex = startIndex + duration)
max_slices <- map2(combos$startIndex, combos$endIndex, function(startIndex, endIndex){
slice(base, startIndex, endIndex) %>%
select(val) %>%
max()
}) %>%
as.numeric()
result <- combos %>%
cbind(max = max_slices)
Result:
startIndex duration endIndex max
1 2 1 3 13
2 4 1 5 15
3 7 1 8 18
4 9 1 10 20
5 2 2 4 14
6 4 2 6 16
7 7 2 9 19
8 9 2 11 19
9 2 3 5 15
10 4 3 7 17
11 7 3 10 20
12 9 3 12 19

Answer 3

I have a solution using the map function however I don't think I have kept the function as a data.table so this may not be satisfactory. Please let me know if not and I can take another look or refer to another answer. One option would be to run the data.table function on the output.
library(tidyverse)
library(data.table)
library(dtplyr)
base <- data.table(idx=c(1,2,3,4,5,6,7,8,9,10), val=c(11,12,13,14,15,16,17,18,19,20))
startIndex <- c(2, 4, 7, 9)
duration <- c(1,2,3)
crossing(startIndex, duration) %>%
data.table() %>%
mutate(max = map2_dbl(startIndex, duration, ~max(base$val[.x:(.x + .y)])))
#> Source: local data table [12 x 3]
#> Call: copy(`_DT1`)[, `:=`(max = map2_dbl(startIndex, duration, ~max(..base$val[.x:(.x +
#> .y)])))]
#>
#> startIndex duration max
#> <dbl> <dbl> <dbl>
#> 1 2 1 13
#> 2 2 2 14
#> 3 2 3 15
#> 4 4 1 15
#> 5 4 2 16
#> 6 4 3 17
#> # ... with 6 more rows
#>
#> # Use as.data.table()/as.data.frame()/as_tibble() to access results
Created on 2021-04-04 by the reprex package (v2.0.0)