Survfit had a bug in some prior releases due to the use of both
unique(times) and table(times); I fixed it by rounding to 15 digits per
the manual page for as.character. Yes, I should ferret out all the
usages instead, but this was fast and it cured the user's problem.
The bug is back! A data set from a local colleage triggers it.
I can send the rda file to anyone who wishes.
The current code has
digits <- floor((.Machine$double.digits) *
logb(.Machine$double.base,10)) #base 10 digits
Y[,1] <- signif(Y[,1], digits)
which gives 15 digits; should I subtract one more?
Should the documentation change?
In the meantime I'm looking at the more permanent fix of turning time
into a factor, then back at the very end. Because it is a bigger change
the potential for breakage is higer, however.
Terry T.
tmt45% R --vanilla
R version 2.12.2 (2011-02-25)
Copyright (C) 2011 The R Foundation for Statistical Computing
ISBN 3-900051-07-0
Platform: x86_64-unknown-linux-gnu (64-bit)
> load('test.rda')
> ls()
[1] "temp2"> temp2 <- round(temp2, 15)
> length(unique(temp2))
[1] 954> length(table(temp2))
[1] 942> .Machine$double.eps
[1] 2.220446e-16> range(temp2)
[1] 0.0000 26.0397
Terry T.
On Mon, Mar 21, 2011 at 10:46:41AM -0500, Terry Therneau wrote:> Survfit had a bug in some prior releases due to the use of both > unique(times) and table(times); I fixed it by rounding to 15 digits per > the manual page for as.character. Yes, I should ferret out all the > usages instead, but this was fast and it cured the user's problem. > The bug is back! A data set from a local colleage triggers it. > I can send the rda file to anyone who wishes. > > The current code has > digits <- floor((.Machine$double.digits) * > logb(.Machine$double.base,10)) #base 10 digits > Y[,1] <- signif(Y[,1], digits) > > which gives 15 digits; should I subtract one more? > > Should the documentation change? > > In the meantime I'm looking at the more permanent fix of turning time > into a factor, then back at the very end. Because it is a bigger change > the potential for breakage is higer, however.Can you describe the error in more detail? Is it related to consistency of converting a number to character and back? Petr Savicky.