separate row containing two separate dates into before and after midnight - r

Question

I have a data frame containing sleep data, with several sleep increments, with a column for the start and a column for the end of the sleep.
For some rows, the starting time is on the previous day and the end time is on the next day.
What I would like to do is to separate such rows into two rows, where the first row contains the starting time till 23:59:59, and the second row 00:00:00 till the end time.
For example:
# A tibble: 6 x 3
sleepdatestarttime sleepdateendtime sleepstage
<dttm> <dttm> <chr>
1 2018-03-02 23:31:00 2018-03-02 23:54:00 rem
2 2018-03-02 23:54:00 2018-03-02 23:55:00 light
3 2018-03-02 23:55:00 2018-03-03 00:02:00 wake
4 2018-03-03 00:02:00 2018-03-03 00:03:30 light
5 2018-03-03 00:03:30 2018-03-03 00:23:30 deep
6 2018-03-03 00:23:30 2018-03-03 02:58:00 light
and the desired output is:
# A tibble: 6 x 3
sleepdatestarttime sleepdateendtime sleepstage
<dttm> <dttm> <chr>
1 2018-03-02 23:31:00 2018-03-02 23:54:00 rem
2 2018-03-02 23:54:00 2018-03-02 23:55:00 light
**3 2018-03-02 23:55:00 2018-03-02 23:59:59 wake
4 2018-03-03 00:00:00 2018-03-03 00:01:59 wake**
5 2018-03-03 00:02:00 2018-03-03 00:03:30 light
6 2018-03-03 00:03:30 2018-03-03 00:23:30 deep
7 2018-03-03 00:23:30 2018-03-03 02:58:00 light
A dplyr solution would be very helpful.

Answer 1

Here is a possible solution but using just base R and not a dplyr. I converted all times to UTC to avoid issue with time conversions. (See a related answer change time zone in R without it returning to original time zone)
Note this solution resorts the entire dataframe by sleepdatestarttime so if there are multiple people on the same day, then the order function on the last line needs modification.
df<-read.table(header=TRUE, text="sleepdatestarttime sleepdateendtime sleepstage
'2018-03-02 23:31:00' '2018-03-02 23:54:00' rem
'2018-03-02 23:54:00' '2018-03-02 23:55:00' light
'2018-03-02 23:55:00' '2018-03-03 00:02:00' wake
'2018-03-03 00:02:00' '2018-03-03 00:03:30' light
'2018-03-03 00:03:30' '2018-03-03 00:23:30' deep
'2018-03-03 00:23:30' '2018-03-03 02:58:00' light")
df$sleepdatestarttime<-as.POSIXct(as.character(df$sleepdatestarttime), tz="UTC")
df$sleepdateendtime<-as.POSIXct(as.character(df$sleepdateendtime), tz="UTC")
#find rows across days
rows<-which(as.Date(df$sleepdatestarttime) !=as.Date(df$sleepdateendtime))
#create the new rows
nstart<-data.frame(sleepdatestarttime= df$sleepdatestarttime[rows],
sleepdateendtime= as.POSIXct(paste(as.Date(df$sleepdatestarttime[rows]), "23:59:59"), tz="UTC"),
sleepstage=df$sleepstage[rows])
nend<-data.frame(sleepdatestarttime= as.POSIXct(paste(as.Date(df$sleepdateendtime[rows]), "00:00:00"), tz="UTC"),
sleepdateendtime= df$sleepdateendtime[rows],
sleepstage=df$sleepstage[rows])
#substitute in the new start rows
df[rows,]<-nstart
#tack on the new ending rows
df<-rbind(df, nend)
#resort the dataframe
df<-df[order(df$sleepdatestarttime ),]

Answer 2

This is a common issue in genomics. The IRanges package on BioConductor has the findOverlaps() function for this purpose. foverlaps() is its data.table version which is used here. AFAIK, there is no dplyr equivalent available.
First we need to create a vector of day start and end times. The call to foverlaps() returns all possible types of overlaps. Finally, the start and end times are adjusted to match with the expected result.
library(data.table)
library(lubridate)
day_seq <- setDT(df)[, .(day_start = seq(
floor_date(min(sleepdatestarttime), "day"),
ceiling_date(max(sleepdateendtime), "day"), "day"))][
, day_end := day_start + days(1)]
setkey(day_seq, day_start, day_end)
foverlaps(
df, day_seq, by.x = c("sleepdatestarttime", "sleepdateendtime"), nomatch = 0L)[
, `:=`(sleepdatestarttime = pmax(sleepdatestarttime, day_start),
sleepdateendtime = pmin(sleepdateendtime, day_end - seconds(1)))][
, c("day_start", "day_end") := NULL][]
i sleepdatestarttime sleepdateendtime sleepstage
1: 1 2018-03-02 23:31:00 2018-03-02 23:54:00 rem
2: 2 2018-03-02 23:54:00 2018-03-02 23:55:00 light
3: 3 2018-03-02 23:55:00 2018-03-02 23:59:59 wake
4: 3 2018-03-03 00:00:00 2018-03-03 00:02:00 wake
5: 4 2018-03-03 00:02:00 2018-03-03 00:03:30 light
6: 5 2018-03-03 00:03:30 2018-03-03 00:23:30 deep
7: 6 2018-03-03 00:23:30 2018-03-03 02:58:00 light
Data
df <- readr::read_table("i sleepdatestarttime sleepdateendtime sleepstage
1 2018-03-02 23:31:00 2018-03-02 23:54:00 rem
2 2018-03-02 23:54:00 2018-03-02 23:55:00 light
3 2018-03-02 23:55:00 2018-03-03 00:02:00 wake
4 2018-03-03 00:02:00 2018-03-03 00:03:30 light
5 2018-03-03 00:03:30 2018-03-03 00:23:30 deep
6 2018-03-03 00:23:30 2018-03-03 02:58:00 light")