>>>>> "BDR" == Prof Brian Ripley <[EMAIL PROTECTED]>
>>>>> on Thu, 10 Jun 2004 08:04:17 +0100 (BST) writes:
BDR> This is surely a bug in jitter, which should pass through a 0-length input
BDR> unchanged as S does (and I will fix that).
BDR> I do wonder what jitter should do with a length-1 input. It does change
BDR> it, using a rather arbitrary notion of the scale of the change (as does
BDR> S). Given the description
BDR> 'jitter(x,...)' returns a numeric of the same length as 'x', but
BDR> with an 'amount' of noise added in order to break ties.
BDR> it appears it should not change a length-1 input.
BDR> Opinions?
I agree with you: It shouldn't change a length-1 input by
default (when no 'amount' is specified)
One could argue that it should add jitter when a positive
'amount' is specified explicitly. But I'd still vote for having
return length-1 arguments unaltered.
Martin
______________________________________________
[EMAIL PROTECTED] mailing list
https://www.stat.math.ethz.ch/mailman/listinfo/r-devel
