[R] Reduced Numerical Precision in 1.2.x in pre-compiled Windows binaries?

Jeff_Lischer@mksinst.com Jeff_Lischer at mksinst.com
Wed Jan 31 18:28:43 CET 2001

On Wed, 31 Jan 2001, Prof Brian D Ripley wrote:

>> I have noticed a definite loss of numerical precision going from 1.1.1
>> 1.2.x under both Windows 98 and NT4 on x86 architecture.

> Did you use exactly the same compiler?  If you used the pre-compiled
> versions you are almost certainly attributing this to the wrong cause,
> so please compile it yourself before making such statements, and do
> check that you have read the CHANGES file.

Thank you for your response to my message. You are correct that I was using
the pre-compiled versions available on the CRAN site. Unfortunately, I
don't have the ability to compile the code myself so I rely on these
pre-compiled versions. What do you think is the real cause of the problem
in the pre-compiled binaries?

Rereading my previous message, I see that I should have chosen my words
more carefully -- I did not mean to imply that R itself had lost precision.
I assumed it was some interaction with the compiler/compiler options and I
thought that this information might be of interest to others on this list.
I have changed the subject line to better reflect the issue.

The only relevant information I found in the CHANGES file was that Versions
1.2.x were compiled against msvcrt.dll not crtdll.dll. The file says that
this change "generally" gives more accurate results.

> I find the code in package boot to be a great deal more trustworthy than
> that in bootstrap.

I tried out the boot and bootstrap packages for this analysis, but I
observed similar issues with the calculated confidence limits. I even tried
going back to the 1.1.1 binaries for both packages, but still saw the
problem. It was easier to determine what was causing the problem in the
bootstrap package because more intermediate parameters are available for
output. Once I determined what was happening, I saw that the troublesome
line is virtually identical in the two packages:

   - In abcnon( ) from bootstrap: t..[i] <- (tp - 2 * t0 + tm)/ep^2
   - In abc.ci( ) from boot: L2[i] <- (t1 - 2 * t0 + t2)/eps^2

Thank you for the recommendation, though. In the future I will start with
the boot package.

D. Jeffrey Lischer, Ph.D.
Principal Mechanical Engineer
MKS Instruments, Inc.
E-Mail: Jeff_Lischer at mksinst.com

r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch

More information about the R-help mailing list