Summarise data between given value of a categorical variable - r

Question

I am looking for a clever and fast way to summarise data in a data frame. The data and desired output looks as follows:
categoriesVector <- c("A", "A", "B", "A", "B", "B", "B", "A", "B")
propertyVector <- 1:length(categoriesVector)
dataVector <- 100 * rev(propertyVector)
df <- data.frame(categoriesVector, propertyVector, dataVector, stringsAsFactors = F)
df
desiredData <- c(700, sum(500, 400, 300), 100)
desiredProperty1 <- c(3, 5, 9)
desiredProperty2 <- c(3, 7, 9)
desiredDF <- data.frame(desiredData, desiredProperty1, desiredProperty2)
desiredDF
Basically I need to sum data and keep first and last property between each two occurrences of Category A. After a lot of headbanging I found a clumsy solution, which I am looking to find an improvement on in terms of clarity and performance, preferably with dplyr:
numRows <- dim(df)[1]
.groupedID <- rep(NA, numRows)
ID <- 1
.groupedID[[1]] <- ifelse(df$categoriesVector[[1]] == "A", 0, ID)
for(i in 2:numRows)
{
if(df$categoriesVector[i] == "B")
{
.groupedID[i] <- ID
if(df$categoriesVector[i - 1] == "B")
{
.groupedID[i] <- .groupedID[i - 1]
}
ID <- ID + 1
} else
{
.groupedID[i] <- 0
}
}
tempDF <-
df %>%
mutate(ID = .groupedID) %>%
filter(ID != 0) %>%
group_by(ID) %>%
summarise(desiredProperty1 = head(propertyVector, 1),
desiredProperty2 = tail(propertyVector, 1),
desiredData = sum(dataVector)) %>%
select(desiredData, desiredProperty1, desiredProperty2)
tempDF

Answer 1

You could use cumsum() to make your groupings and then process based on those like this:
df %>% mutate(Agroups = cumsum(categoriesVector == "A")) %>%
filter(categoriesVector == "B") %>%
group_by(Agroups) %>%
summarise(propertyStart = min(propertyVector),
propertyEnd = max(propertyVector),
dataTotal = sum(dataVector))
# A tibble: 3 x 4
Agroups propertyStart propertyEnd dataTotal
<int> <dbl> <dbl> <dbl>
1 2 3 3 700
2 3 5 7 1200
3 4 9 9 100

Answer 2

Here is how I'd do with data.table. First create spanNumber variable to identify each span of "B" surrounded by "A", then calculate the variables you specified:
library(data.table)
setDT(df)
df[, catShiftConcat := paste0(categoriesVector, shift(categoriesVector, fill = "A"))]
df[categoriesVector == "B", spanNumber := cumsum(catShiftConcat == "BA")]
df[!is.na(spanNumber) , .(desiredData = sum(dataVector),
desiredProperty1 = propertyVector[1],
desiredProperty2 = propertyVector[.N]), by = spanNumber]
## spanNumber desiredData desiredProperty1 desiredProperty2
## 1: 1 700 3 3
## 2: 2 1200 5 7
## 3: 3 100 9 9

Answer 3

An alternative data.table method that uses rleid to group runs of the categories vector is
library(data.table)
setDT(df)[, .(categoriesVector,
desiredData=sum(dataVector),
desiredProperty1=propertyVector[1],
desiredProperty2=propertyVector[.N]),
by=rleid(categoriesVector)
][categoriesVector == "B",][, c("rleid", "categoriesVector") := NULL][]
The contents in the first [] return the desired output and are calculated aggregated to runs of the categories vector. The second chain subsets the observations by keeping those for which the categories vector is B. The third [] removes two helper variables, and the final [] is just there to print the result to screen.
This returns
desiredData desiredProperty1 desiredProperty2
1: 700 3 3
2: 1200 5 7
3: 1200 5 7
4: 1200 5 7
5: 100 9 9