Breaking down a timed sequence into episodes - r

Question

I'm trying to break down a vector of event times into episodes. An episode must meet 2 criteria. 1) It consists of 3 or more events and 2) those events have inter-event times of 25 time units or less. My data is organized in a data frame as shown below.
So far, I figured out that I can find the difference between events with diff(EventTime). By creating a logical vector that corresponds to events that the 2nd inter-event criterion, I can use rle(EpisodeTimeCriterion) to get a the total number, and length of episodes.
EventTime TimeDifferenceBetweenNextEvent EpisodeTimeCriterion
25 NA NA
75 50 TRUE
100 25 TRUE
101 1 TRUE
105 4 TRUE
157 52 FALSE
158 1 TRUE
160 2 TRUE
167 7 TRUE
169 2 TRUE
170 1 TRUE
175 5 TRUE
178 3 TRUE
278 100 FALSE
302 24 TRUE
308 6 TRUE
320 12 TRUE
322 459 FALSE
However, I would like to know the timing of the episodes and 'rle()' doesn’t let me do that.
Ideally I would like to generate a data frame that looks like this:
Episode EventsPerEpisode EpisodeStartTime EpisodeEndTime
1 4 75 105
2 7 158 178
3 3 302 322
I know that this is probably a simple problem but being new to R, the only solution I can envision is some series of loops. Is there a way of doing this without loops? Or is there package that lends itself to this sort of analysis?
Thanks!
Edited for clarity. Added a desired outcome data fame and expanded the example data to make it clearer.

Answer 1

You've got the pieces you need. You really need to make a variable that gives each episode a number/name so you can group by it. rle(...)$length gives you run lengths, so just use rep to repeat a number that number of times:
runs <- rle(df$EpisodeTimeCriterion)$lengths # You don't need this extra variable, but it makes the code more readable
df$Episode <- rep(1:length(runs), runs)
so df looks like
> head(df)
EventTime TimeDifferenceBetweenNextEvent EpisodeTimeCriterion Episode
1 25 NA NA 1
2 75 50 TRUE 2
3 100 25 TRUE 2
4 101 1 TRUE 2
5 105 4 TRUE 2
6 157 52 FALSE 3
Now use dplyr to summarize the data:
library(dplyr)
df2 <- df %>% filter(EpisodeTimeCriterion) %>% group_by(Episode) %>%
summarise(EventsPerEpisode = n(),
EpisodeStartTime = min(EventTime),
EpisodeEndTime = max(EventTime))
which returns
> df2
Source: local data frame [3 x 4]
Episode EventsPerEpisode EpisodeStartTime EpisodeEndTime
(int) (int) (dbl) (dbl)
1 2 4 75 105
2 4 7 158 178
3 6 3 302 320
If you want your episode numbers to be integers starting with one, you can clean up with
df2$Episode <- 1:nrow(df2)
Data
If someone wants to play with the data, the results of dput(df) before running the above code:
df <- structure(list(EventTime = c(25, 75, 100, 101, 105, 157, 158,
160, 167, 169, 170, 175, 178, 278, 302, 308, 320, 322), TimeDifferenceBetweenNextEvent = c(NA,
50, 25, 1, 4, 52, 1, 2, 7, 2, 1, 5, 3, 100, 24, 6, 12, 459),
EpisodeTimeCriterion = c(NA, TRUE, TRUE, TRUE, TRUE, FALSE,
TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, FALSE, TRUE, TRUE,
TRUE, FALSE)), .Names = c("EventTime", "TimeDifferenceBetweenNextEvent",
"EpisodeTimeCriterion"), row.names = c(NA, -18L), class = "data.frame")

Answer 2

Here is one approach I came up with using a combination of cut2 from Hmisc package and cumsum to label episodes into numbers:
library(Hmisc)
library(dplyr)
df$episodeCut <- cut2(df$TimeDifferenceBetweenNextEvent, c(26))
df$episode <- cumsum((df$episodeCut == '[ 1,26)' & lag(df$episodeCut) != '[ 1,26)') | df$episodeCut != '[ 1,26)')
Output is as follows:
EventTime TimeDifferenceBetweenNextEvent EpisodeTimeCriterion episodeCut episode
1 25 50 FALSE [26,52] 1
2 75 25 TRUE [ 1,26) 2
3 100 1 TRUE [ 1,26) 2
4 101 4 TRUE [ 1,26) 2
5 105 52 TRUE [26,52] 3
6 157 52 FALSE [26,52] 4
As you can see, it tags rows 2, 3, 4 as belonging to a single episode.
Is this what you are looking for? Not sure from your description. So, my answer may be wrong.