Skip to content

Latest commit

 

History

History
152 lines (121 loc) · 3.69 KB

benchmark.md

File metadata and controls

152 lines (121 loc) · 3.69 KB
Raw

Benchmarking timeless and anytime

library(timeless)
library(anytime)

The benchmark is done with three datasets that contain a variety of different date(time) formats.

bench_datetimes <- readLines("datetime1000.txt")
head(bench_datetimes)
[1] "28 December 1979 12:54AM" "12/21/1991 08:07 AM"     
[3] "13-03-1979 19:51"         "2007-11-08 01:09:25"     
[5] "May 29, 1978 07:57"       "2015-05-04 21:55:07"     

bench_epochs <- readLines("epoch500.txt")
head(bench_epochs)
[1] "717700128" "115153946" "948771719" "586380132" "795097964" "211051179"

bench_dates <- readLines("dates500.txt")
head(bench_dates)
[1] "September 07, 2018"         "1991.02.14"                
[3] "12:00 AM December 26, 2000" "April 05, 1996"            
[5] "June 19, 2014"              "27-Jun-2016"               

bench <- c(bench_datetimes, bench_epochs, bench_dates)

Ability to parse

This benchmark just checks if something was parsed and does not say if the result is actually correct.

sum_na <- function(x) sum(is.na(x))
data.frame(
    type = c("datetimes", "epochs", "dates", "all"),
    chronos = c(
        sum_na(chronos(bench_datetimes)),
        sum_na(chronos(bench_epochs)),
        sum_na(chronos(bench_dates, out_format = "date")),
        sum_na(chronos(bench))
    ),
    anytime = c(
        sum_na(anytime(bench_datetimes)),
        sum_na(anytime(as.numeric(bench_epochs))),
        sum_na(anydate(bench_dates)),
        sum_na(anytime(bench))
    )
)
       type chronos anytime
1 datetimes       0     322
2    epochs       0       0
3     dates       0     138
4       all       0     949

When epoch times are encoded as characters (which happens when all data is put together in one vector), then anytime fails to parse most of the epoch times.

Runtime

The package fasttime can be used together with timeless to convert larger sets of datetimes by letting chronos return a character vector which is then parsed by fastPOSIXct.

fast_chronos <- function(x, out_format = "datetime") {
    res <- chronos(x, out_format = "character")
    if (out_format == "datetime") {
        return(fasttime::fastPOSIXct(res))
    } else {
        return(fasttime::fastDate(res))
    }
}

Full data

mb <- microbenchmark::microbenchmark(
    chronos = chronos(bench),
    fast_chronos = fast_chronos(bench),
    anytime = anytime(bench),
    times = 100L
)
ggplot2::autoplot(mb)

datetime

mb <- microbenchmark::microbenchmark(
    chronos = chronos(bench_datetimes),
    fast_chronos = fast_chronos(bench_datetimes),
    anytime = anytime(bench_datetimes),
    times = 100L
)
ggplot2::autoplot(mb)

epoch

bench_epochs_num <- as.integer(bench_epochs)
mb <- microbenchmark::microbenchmark(
    chronos = chronos(bench_epochs_num),
    fast_chronos = fast_chronos(bench_epochs_num),
    anytime = anytime(bench_epochs_num),
    posix = as.POSIXct(bench_epochs_num),
    fastposix = fasttime::fastPOSIXct(bench_epochs_num),
    times = 100L
)
ggplot2::autoplot(mb)

When the date vector only consists of epoch timestamps, it is best to parse them diretly with as.POSIXct

date

mb <- microbenchmark::microbenchmark(
    chronos = chronos(bench_date, out_format = "date"),
    fast_chronos = fast_chronos(bench_date, out_format = "date"),
    anytime = anydate(bench_date),
    times = 100L
)
ggplot2::autoplot(mb)