R help - Apr 2013 - Floating point precision causing undesireable behaviour when printing as.POSIXlt times with microseconds?

Dear list,
When using as.POSIXlt with times measured down to microseconds the default
format.POSIXlt seems to cause some possibly undesirable behaviour:

According to the code in format.POSIXlt the maximum accuracy of printing
fractional seconds is 1 microsecond, but if I do;

options( digits.secs = 6 )
as.POSIXlt( 1.000002 , tz="", origin="1970-01-01")
as.POSIXlt( 1.999998 , tz="", origin="1970-01-01")
as.POSIXlt( 1.999999 , tz="", origin="1970-01-01")

I return respectively:
[1] "1970-01-01 01:00:01.000002 BST"
[1] "1970-01-01 01:00:01.999998 BST"
[1] "1970-01-01 01:00:01 BST"

If options( digits.secs = 6 ) should I not expect to be able to print 1.999999
seconds? This seems to be caused by the following code fragment in
format.POSIXlt:

np <- getOption("digits.secs")
        if (is.null(np))
            np <- 0L
        else np <- min(6L, np)
        if (np >= 1L)
            for (i in seq_len(np) - 1L) if (all(abs(secs - round(secs,
                i)) < 1e-06)) {
                np <- i
                break
            }

Specifically
            for (i in seq_len(np) - 1L) if (all(abs(secs - round(secs,
                i)) < 1e-06))
Which in the case of 1.999999 seconds will give:

options( scipen = 10 )
np <- 6
sapply( seq_len(np) - 1L , function(x) abs(1.999999 - round(1.999999, x)) )

         [,1]     [,2]     [,3]     [,4]     [,5]     [,6]
[1,] 0.000001 0.000001 0.000001 0.000001 0.000001 0.000001

The logical test all( ... < 1e-06)  should evaluate to FALSE but due to
floating point precision it evaluates TRUE:

sprintf( "%.20f" , abs(1. 999999  - round(1. 999999,5)))
[1] "0.00000099999999991773"

If instead of:

            for (i in seq_len(np) - 1L) if (all(abs(secs - round(secs,
                i)) < 1e-06))

in format.POSIXlt we had a comparison value that was half the minimum increment:

            for (i in seq_len(np) - 1L) if (all(abs(secs - round(secs,
                i)) < 5e-07))

This behaviour disappears:

mod.format.POSIXlt( as.POSIXlt( 1.999999 , tz="",
origin="1970-01-01") )
[1] "1970-01-01 01:00:01.999999"

But I am unsure if the original behaviour is what I should expect given the
documentation (I have read it and I can't see a reason to expect 1.999999 to
round down to 1). And also if changing the formatting function would have other
undesirable consequences?

My sessionInfo():
R version 3.0.0 (2013-04-03)
Platform: x86_64-w64-mingw32/x64 (64-bit)

locale:
[1] LC_COLLATE=English_United Kingdom.1252
[2] LC_CTYPE=English_United Kingdom.1252
[3] LC_MONETARY=English_United Kingdom.1252
[4] LC_NUMERIC=C
[5] LC_TIME=English_United Kingdom.1252

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base



Thank you,


Simon


------------------------------------
Simon O'Hanlon
Postgraduate Researcher

Helminth Ecology Research Group
Department of Infectious Disease Epidemiology
Imperial College London
St. Mary's Hospital, Norfolk Place,
London, W2 1PG, UK

Office: +44 (0) 20 759 43229
------------------------------------


	[[alternative HTML version deleted]]

FAQ 7.31

also if you are using POSIXct for current dates, the resolution is down to about
a milliseconds.

Sent from my iPad

On Apr 24, 2013, at 13:57, "O'Hanlon, Simon J" <simon.ohanlon
at imperial.ac.uk> wrote:
> Dear list,
> When using as.POSIXlt with times measured down to microseconds the default
format.POSIXlt seems to cause some possibly undesirable behaviour:
> 
> According to the code in format.POSIXlt the maximum accuracy of printing
fractional seconds is 1 microsecond, but if I do;
> 
> options( digits.secs = 6 )
> as.POSIXlt( 1.000002 , tz="", origin="1970-01-01")
> as.POSIXlt( 1.999998 , tz="", origin="1970-01-01")
> as.POSIXlt( 1.999999 , tz="", origin="1970-01-01")
> 
> I return respectively:
> [1] "1970-01-01 01:00:01.000002 BST"
> [1] "1970-01-01 01:00:01.999998 BST"
> [1] "1970-01-01 01:00:01 BST"
> 
> If options( digits.secs = 6 ) should I not expect to be able to print
1.999999 seconds? This seems to be caused by the following code fragment in
format.POSIXlt:
> 
> np <- getOption("digits.secs")
>        if (is.null(np))
>            np <- 0L
>        else np <- min(6L, np)
>        if (np >= 1L)
>            for (i in seq_len(np) - 1L) if (all(abs(secs - round(secs,
>                i)) < 1e-06)) {
>                np <- i
>                break
>            }
> 
> Specifically
>            for (i in seq_len(np) - 1L) if (all(abs(secs - round(secs,
>                i)) < 1e-06))
> Which in the case of 1.999999 seconds will give:
> 
> options( scipen = 10 )
> np <- 6
> sapply( seq_len(np) - 1L , function(x) abs(1.999999 - round(1.999999, x)) )
> 
>         [,1]     [,2]     [,3]     [,4]     [,5]     [,6]
> [1,] 0.000001 0.000001 0.000001 0.000001 0.000001 0.000001
> 
> The logical test all( ... < 1e-06)  should evaluate to FALSE but due to
floating point precision it evaluates TRUE:
> 
> sprintf( "%.20f" , abs(1. 999999  - round(1. 999999,5)))
> [1] "0.00000099999999991773"
> 
> If instead of:
> 
>            for (i in seq_len(np) - 1L) if (all(abs(secs - round(secs,
>                i)) < 1e-06))
> 
> in format.POSIXlt we had a comparison value that was half the minimum
increment:
> 
>            for (i in seq_len(np) - 1L) if (all(abs(secs - round(secs,
>                i)) < 5e-07))
> 
> This behaviour disappears:
> 
> mod.format.POSIXlt( as.POSIXlt( 1.999999 , tz="",
origin="1970-01-01") )
> [1] "1970-01-01 01:00:01.999999"
> 
> But I am unsure if the original behaviour is what I should expect given the
documentation (I have read it and I can't see a reason to expect 1.999999 to
round down to 1). And also if changing the formatting function would have other
undesirable consequences?
> 
> My sessionInfo():
> R version 3.0.0 (2013-04-03)
> Platform: x86_64-w64-mingw32/x64 (64-bit)
> 
> locale:
> [1] LC_COLLATE=English_United Kingdom.1252
> [2] LC_CTYPE=English_United Kingdom.1252
> [3] LC_MONETARY=English_United Kingdom.1252
> [4] LC_NUMERIC=C
> [5] LC_TIME=English_United Kingdom.1252
> 
> attached base packages:
> [1] stats     graphics  grDevices utils     datasets  methods   base
> 
> 
> 
> Thank you,
> 
> 
> Simon
> 
> 
> ------------------------------------
> Simon O'Hanlon
> Postgraduate Researcher
> 
> Helminth Ecology Research Group
> Department of Infectious Disease Epidemiology
> Imperial College London
> St. Mary's Hospital, Norfolk Place,
> London, W2 1PG, UK
> 
> Office: +44 (0) 20 759 43229
> ------------------------------------
> 
> 
>    [[alternative HTML version deleted]]
> 
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.