From: Petr Savicky <savicky_at_cs.cas.cz>

Date: Mon, 03 Sep 2007 16:39:57 +0200

R-devel_at_r-project.org mailing list

https://stat.ethz.ch/mailman/listinfo/r-devel Received on Tue 04 Sep 2007 - 04:56:13 GMT

Date: Mon, 03 Sep 2007 16:39:57 +0200

On Mon, Sep 03, 2007 at 08:59:22AM +0200, marco.vicentini_at_gmail.com wrote:

> Full_Name: Marco Vicentini, University of Verona

*> Version: 2.4.1 & 2.5.1
**> OS: OsX & WinXP
**> Submission from: (NULL) (157.27.253.46)
**>
**>
**> When I proceed to test the following equation 1 + 2 == 3, I obviously obtain the
**> value TRUE. But when I tryed to do the same using real number (i.e. 0.1 + 0.2 ==
**> 0.3) I obtained an unusual FALSE.
**> In the online help there are some tricks for this problem. It suggests to use
**> identical(...) which again answer FALSE. Only using isTRUE(all.equal(0.3, 0.1 +
**> 0.2)) I can obtain the true value TRUE.
*

A rational number has a finite binary expansion iff its denominator is a power
of 2. Numbers 0.1 and 0.2 are 1/10 and 1/5, so they have 5 in their denominator.
Their binary expansion is

0.1 = .0001100110011001100110011001100110...
0.2 = .0011001100110011001100110011001100...
A double variable stores the numbers rounded to 53 significant binary digits.
Hence, they are not exactly 0.1 and 0.2, as may be seen in

formatC(0.1,digits=30) # [1] "0.100000000000000005551115123126"formatC(0.2,digits=30) # [1] "0.200000000000000011102230246252"

In order to compare numbers with some tolerance, the function all.equal may be used, which you also mention below. See its help page, which specifies the tolerance to be .Machine$double.eps ^ 0.5.

> But the problem does not concern only the operator ==. Many other functions,

*> among over: sort, order, unique, duplicate, identical are not able to deal with
**> this problem. This is very dangerous because no advice are provide by the online
**> help, and anybody can use these functions no think to unusual results.
**>
**> I think that the problem is due to how double number are store by the C
**> compiler.
*

Not C compiler, but the hardware.

Petr Savicky.

> If it may be usefull, I have written to small function (Unique and isEqual)

*> which can deal with this problem of the double numbers.
**>
**> I also add some other conditions for the same problem.
**>
**> 0.3 == 0.15 + 0.15
**> 0.3 == 0.1 + 0.2
**> 1 - 0.7 == 0.3
**> 0.1 == 1 - 0.9
**>
**> 0.2 == 1 - 0.2 - 0.2 - 0.2 - 0.2
**> -0.2 == 1 - 0.2 - 0.2 - 0.2 - 0.2 - 0.2 - 0.2
**>
**> identical (0.3, 0.1 + 0.2)
**> all.equal (0.3, 0.1 + 0.2)
**>
**> identical (-0.2 , 1 - 0.2 - 0.2 - 0.2 - 0.2 - 0.2 - 0.2)
**> all.equal (-0.2 , 1 - 0.2 - 0.2 - 0.2 - 0.2 - 0.2 - 0.2)
**>
**> isTRUE( all.equal (-0.2 , 1 - 0.2 - 0.2 - 0.2 - 0.2 - 0.2 - 0.2) )
**>
**>
**> -0.2 == 1 - 0.2 - 0.2 - 0.2 - 0.2 - 0.2 - 0.2
**>
**> a= -0.2
**> b= 1 - 0.2 - 0.2 - 0.2 - 0.2 - 0.2 - 0.2
**>
**> x<-c(a,b)
**> sprintf("%.15f",x)
**> sprintf("%.50f",x)
**>
**>
**>
**> Unique <- function(x, digits = 8, fast = TRUE) {
**>
**> if (fast) {
**> unique (round(x * 10^digits)) / 10^digits
**> } else {
**> x = sort(x)
**> for (i in 1:(length(x)-1))
**> if (isTRUE(all.equal(x[i],x[i+1]))) x[i] = NaN
**> x [ which (!is.nan(x)) ]
**> }}
**>
**> isEqual <- function (object, x, tol = 1e-9) {
**> if (!is.vector(object)) stop("Object must be a vector")
**> if (is.character(object)) stop("Object can not be a character")
**> if (!is.real(x)) stop("x must be a real number")
**> if (any(is.na(c(object,x)))) stop("NA is not supported")
**> if (length(x) != 1) stop("length x must equal to 1")
**>
**> ifelse (abs(object - x) < tol, TRUE,FALSE)
**> # .Call("isEqual",as.real(object),as.real(x),as.real(tol), PACKAGE="mvUtils")
**> }
**>
**> ______________________________________________
**> R-devel_at_r-project.org mailing list
**> https://stat.ethz.ch/mailman/listinfo/r-devel
*

>

R-devel_at_r-project.org mailing list

https://stat.ethz.ch/mailman/listinfo/r-devel Received on Tue 04 Sep 2007 - 04:56:13 GMT

Archive maintained by Robert King, hosted by
the discipline of
statistics at the
University of Newcastle,
Australia.

Archive generated by hypermail 2.2.0, at Tue 04 Sep 2007 - 14:40:10 GMT.

*
Mailing list information is available at https://stat.ethz.ch/mailman/listinfo/r-devel.
Please read the posting
guide before posting to the list.
*