Rounding time to nearest quarter hour

Question

I have a vector of POSIXct values and I would like to round them to the nearest quarter hour. I don\'t care about the day. How do I convert the values to hours and minutes?<

Answer 1

Indeed, an old question with some helpful answers so far. The last one by giraffhere seems to be the most elegant. however, ist not floor_date but round_date which will do the trick:

lubridate::round_date(x, "15 minutes") 

Answer 2

You can use round. The trick is to divide by 900 seconds (15 minutes * 60 seconds) before rounding and multiply by 900 afterwards:

a <-as.POSIXlt("2012-05-30 20:41:21 UTC")
b <-as.POSIXlt(round(as.double(a)/(15*60))*(15*60),origin=(as.POSIXlt('1970-01-01')))
b
[1] "2012-05-30 20:45:00 EDT"

To get only hour and minute, just use format

format(b,"%H:%M")
[1] "20:45"

as.character(format(b,"%H:%M"))
[1] "20:45"

Answer 3

Try this, which combines both requests and is based on looking at what round.POSIXt() and trunc.POSIXt() do.

myRound <- function (x, convert = TRUE)  {
    x <- as.POSIXlt(x)
    mins <- x$min
    mult <- mins %/% 15
    remain <- mins %% 15
    if(remain > 7L || (remain == 7L && x$sec > 29))
        mult <- mult + 1
    if(mult > 3) {
        x$min <- 0
        x <- x + 3600
    } else {
        x$min <- 15 * mult
    }
    x <- trunc.POSIXt(x, units = "mins")
    if(convert) {
        x <- format(x, format = "%H:%M")
    }
    x
}

This gives:

> tmp <- as.POSIXct("2012-05-30 20:41:21 UTC")
> myRound(tmp)
[1] "20:45"
> myRound(tmp, convert = FALSE)
[1] "2012-05-30 20:45:00 BST"
> tmp2 <- as.POSIXct("2012-05-30 20:55:21 UTC")
> myRound(tmp2)
[1] "21:00"
> myRound(tmp2, convert = FALSE)
[1] "2012-05-30 21:00:00 BST"

Answer 4

Using IDate and ITime classes from data.table and a IPeriod class (just developed) I was able to get more scalable solution.
Only shhhhimhuntingrabbits and PLapointe answer the question in terms of nearest. xts solution only rounds using ceiling, my IPeriod solution allows to specify ceiling or floor.
To get top performance you would need to keep your data in IDate and ITime classes. As seen on benchmark it is cheap to produce POSIXct from IDate/ITime/IPeriod. Below benchmark of some 22M timestamp:

# install only if you don't have
install.packages(c("microbenchmarkCore","data.table"),
                 repos = c("https://olafmersmann.github.io/drat",
                           "https://jangorecki.github.io/drat/iperiod"))
library(microbenchmarkCore)
library(data.table) # iunit branch
library(xts)
Sys.setenv(TZ="UTC")

## some source data: download and unzip csv
# "http://api.bitcoincharts.com/v1/csv/btceUSD.csv.gz"
# below benchmark on btceUSD.csv.gz 11-Oct-2015 11:35 133664801

system.nanotime(dt <- fread(".btceUSD.csv"))
# Read 21931266 rows and 3 (of 3) columns from 0.878 GB file in 00:00:10
#     user   system  elapsed 
#       NA       NA 9.048991

# take the timestamp only
x = as.POSIXct(dt[[1L]], tz="UTC", origin="1970-01-01")

# functions
shhhhi <- function(your.time){
    strptime("1970-01-01", "%Y-%m-%d", tz="UTC") + round(as.numeric(your.time)/900)*900
}

PLapointe <- function(a){
    as.POSIXlt(round(as.double(a)/(15*60))*(15*60),origin=(as.POSIXlt('1970-01-01')))
}

# myRound - not vectorized

# compare results
all.equal(
    format(shhhhi(x),"%H:%M"),
    format(PLapointe(x),"%H:%M")
)
# [1] TRUE
all.equal(
    format(align.time(x, n = 60*15),"%H:%M"),
    format(periodize(x, "mins", 15),"%H:%M")
)
# [1] TRUE

# IPeriod native input are IDate and ITime - will be tested too
idt <- IDateTime(x)
idate <- idt$idate
itime <- idt$itime
microbenchmark(times = 10L,
               shhhhi(x),
               PLapointe(x),
               xts = align.time(x, 15*60),
               posix_ip_posix = as.POSIXct(periodize(x, "mins", 15), tz="UTC"),
               posix_ip = periodize(x, "mins", 15),
               ip_posix = as.POSIXct(periodize(idate, itime, "mins", 15), tz="UTC"),
               ip = periodize(idate, itime, "mins", 15))
# Unit: microseconds
#            expr         min          lq         mean       median          uq         max neval
#       shhhhi(x)  960819.810  984970.363 1127272.6812 1167512.2765 1201770.895 1243706.235    10
#    PLapointe(x) 2322929.313 2440263.122 2617210.4264 2597772.9825 2792936.774 2981499.356    10
#             xts  453409.222  525738.163  581139.6768  546300.9395  677077.650  767609.155    10
#  posix_ip_posix 3314609.993 3499220.920 3641219.0876 3586822.9150 3654548.885 4457614.174    10
#        posix_ip 3010316.462 3066736.299 3157777.2361 3133693.0655 3234307.549 3401388.800    10
#        ip_posix     335.741     380.696     513.7420     543.3425     630.020     663.385    10
#              ip      98.031     151.471     207.7404     231.8200     262.037     278.789    10

IDate and ITime successfully scales not only in this particular task. Both types, same as IPeriod, are integer based. I would assume they will also scale nice on join or grouping by datetime fields.
Online manual: https://jangorecki.github.io/drat/iperiod/

Answer 5

something like

format(strptime("1970-01-01", "%Y-%m-%d", tz="UTC") + round(as.numeric(your.time)/900)*900,"%H:%M")

would work

Answer 6

You can use the align.time function in the xts package to handle the rounding, then format to return a string of "HH:MM":

R> library(xts)
R> p <- as.POSIXct("2012-05-30 20:41:21", tz="UTC")
R> a <- align.time(p, n=60*15)  # n is in seconds
R> format(a, "%H:%M")
[1] "20:45"