Re: density(kernel = "cosine") .. the `wrong cosine' ..

About this list Date view Thread view Subject view Author view Other groups

Subject: Re: density(kernel = "cosine") .. the `wrong cosine' ..
From: Martin Maechler (
Date: Thu 02 Dec 1999 - 00:47:51 EST

Message-ID: <14405.13591.209149.728895@gargle.gargle.HOWL>

>>>>> "BDR" == Prof Brian D Ripley <> writes:

BDR> On Wed, 1 Dec 1999, Martin Maechler wrote:
>> Problem:
>> - An average user knowing some statistics literature will most
>> probably assume that a "cosine" kernel means the one in the
>> literature, *NOT* the one we have in R now.

BDR> Or they have know what is in S (or what V&R say it is).
BDR> (Yes, we knew of the discrepancy, so defined it.)

I've looked in MASS 3rd ed., before my post; but didn't find that remark...
(now I did : middle of p.134).

But then, our density() is not S compatible
(and the help page has always said so) anyway,
and doesn't want to copy
some of that sillynesses (default bandwidth; gaussian factor of 2 [or 4?]),..

>> Proposition / Possibilities / RFC [= Request For Comments] :
>> - We CHANGE the behavior of density(* , kernel="cosine") to use the
>> cosine from the litterature.

BDR> I am against that. S compatibility and all that.

>> - provide the current "cosine" as kernel = "smoothcosine" {I'd like
>> to keep the possibility of 1-initial-letter abbreviation}

BDR> OK, or

BDR> 3) As it is confusing and never used(?), drop it altogether.

that's against S compatibility as well.
*AND* we make it `impossible' to do what was possible earlier.

>> Enhancement (easy, I'll do that):
>> - We further provide both Epanechnikov and "quartic" aka "biweight"
>> additionally in any case.

BDR> You may find it hard to get agreement on what those are (the
BDR> problem being the scale factors).

That shouldn't matter as long as I provide an `equivalent bandwidth'
factor, should it?
I think nobody can assume that something like
   ``density with Epanechnikov kernel & bandwidth 2'' is universally well
defined. But one should be able to easily compute an Epanechnikov kernel
estimate -- automatically using "equivalent bandwidths".

r-devel mailing list -- Read
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !) To:

About this list Date view Thread view Subject view Author view Other groups

This archive was generated by hypermail 2b25 : Tue 04 Jan 2000 - 14:16:11 EST