Faster way to count occurrences in 5 minute segments?

Question

I have a matrix, `events`, that contains the times of occurrences of 5 million events. Each of these 5 million events has a "type" that ranges from 1 to 2000. A very simplified version of the matrix is as below. The units for "times" is seconds since 1970. All of the events have occurred since 1/1/2012. ``` >events type times 1 1352861760 1 1362377700 2 1365491820 2 1368216180 2 1362088800 2 1362377700 ``` I am trying to divide the time since 1/1/2012 into 5-minute buckets and then populate each of these buckets with how many of each event of type `i` has occurred in each bucket. My code is below. Note that `types` is a vector containing each possible type from 1-2000, and `by` is set to 300 because that is how many seconds are in 5 minutes. ``` for(i in 1:length(types)){ local <- events[events$type==types[i],c("type", "times")] assign(sprintf("a%d", i),table(cut(local$times, breaks=seq(range(events$times)[1],range(events$times)[2], by=300)))) } ``` This results in variables `a1` thro…

Accepted Answer

I would use the xts package for this. Running a function over non-overlapping 5-minute intervals is easy with the `period.apply` and `endpoints` functions.

```
# create sample data
library(xts)
set.seed(21)
N <- 1e6
events <- cbind(sample(2000, N, replace=TRUE),
  as.POSIXct("2012-01-01")+sample(1e7,N))
colnames(events) <- c("type","times")
# create xts object
xevents <- xts(events[,"type"], .POSIXct(events[,"times"]))
# find the last row of each non-overlapping 5-minute interval
ep <- endpoints(xevents, "minutes", 5)
# count the number of occurrences of each "type"
counts <- period.apply(xevents, ep, tabulate, nbins=2000)
# set colnames
colnames(counts) <- paste0("a",1:ncol(counts))
# calculate correlation
#cc <- cor(counts)
```

Update to respond to OP's comments/edits:

```
# Create a sequence of 5-minute steps, from the actual start of the data
m5 <- seq(round(start(xevents),'mins'), end(xevents), by='5 mins')
# Create a sequence of 5-minute steps, from the start of 2012-01-01
m5 <- seq(as.POSIXct("2012-01-01"), end(xevents), by='5 mins')
# merge xevents with empty 5-minute xts object, and
# subtract 1 second, so endpoints are at end of each 5-minute interval
xevents5 <- merge(xevents, xts(,m5-1))
ep5 <- endpoints(xevents5, "minutes", 5)
counts5 <- period.apply(xevents5, ep5, tabulate, nbins=2000)
colnames(counts5) <- paste0("a",1:ncol(counts5))
# align to the beginning of each 5-minute interval, if you want
counts5 <- align.time(counts5,60*5)
```

Faster way to count occurrences in 5 minute segments?

Solution

Problem

Related problems