R rounding issue on fractional seconds

Question

I'm trying to create a sequence for a timeseries sampled at a frequency of 10Hz (i.e. a period of 0.1 secs)

Following this post I used:

> options(digits.secs=4)

then:

> time.seq = seq(from=as.POSIXlt("00:00:00.0", format="%H:%M:%OS",tz="GMT"),                   length.out=10, by=0.10)

which returns:

> head(time.seq)
[1] "2018-02-08 00:00:00.0 GMT" "2018-02-08 00:00:00.0 GMT"
[3] "2018-02-08 00:00:00.2 GMT" "2018-02-08 00:00:00.2 GMT"
[5] "2018-02-08 00:00:00.4 GMT" "2018-02-08 00:00:00.5 GMT"

This isn't right. 2 should be:

"2018-02-08 00:00:00.1 GMT"

and [4]

"2018-02-08 00:00:00.3 GMT"

Initially I thought that the problem was with formatting or getOption but that doesn't seem to be the case since if I use a different fraction, it works perfectly:

> time.seq = seq(from=as.POSIXlt("00:00:00.0", format="%H:%M:%OS",tz="GMT"),                   length.out=10, by=0.0315)
> head(time.seq)
[1] "2018-02-08 00:00:00.0000 GMT" "2018-02-08 00:00:00.0315 GMT"
[3] "2018-02-08 00:00:00.0629 GMT" "2018-02-08 00:00:00.0945 GMT"
[5] "2018-02-08 00:00:00.1259 GMT" "2018-02-08 00:00:00.1575 GMT"

So this implies that there is some rounding error.

Does anyone have any ideas how to fix this, please? I'm using R 3.4.1 on OS X (High Sierra)

Thanks!

UPDATE From this post I can see that there this is a floating point error in the addition. Using the 'form' formatting function in that post from Matthew Lundberg I get:

f4 <- "%Y-%m-%d %H:%M:%OS4"
> time.seq = seq(from=as.POSIXlt("00:00:00.0", format="%H:%M:%OS",tz="GMT"),                   length.out=10, by=0.10)
> format(time.seq[1], f4)
[1] "2018-02-09 00:00:00.0000"
> format(time.seq[2], f4)
[1] "2018-02-09 00:00:00.0999"
> format(time.seq[3], f4)
[1] "2018-02-09 00:00:00.2000"

This presumably means that for some reason the formatting in R isn't picking up my option setting:

> options(digits.secs=4)

Computers aren't as good at adding decimals as you think they are. The problem is basically the same as: https://stackoverflow.com/questions/9508518/why-are-these-numbers-not-equal — MrFlick, Feb 08 '18 at 16:33

RolandASc · Answer 1 · 2018-02-12T10:21:29.827

There is indeed an inconsistency between

format.POSIXlt (which is also called by print.POSIXlt) determining the number of necessary relevant digits (considering getOption("digits.secs") as the max, but in your case relevant digits would be just 1 since .099999 rounded to 4 digits is still .1000)
the internal code called by format.POSIXlt as .Internal(format.POSIXlt(x, format, usetz)), which probably just cuts (or floors?) to the determined number of relevant digits, without actually rounding the same way

Edit:
Having had time to look at the linked post how-r-formats-posixct-with-fractional-seconds, I guess everything else has been said / discussed there, in particular, that rounding up could be more involved if you reach 60 seconds, which might be the main reason why things are as they are.

R rounding issue on fractional seconds

1 Answers1