R help - May 2010 - custom metric for dist for use with hclust/kmeans

Hi guys,

I've been using the kmeans and hclust functions for some time now and
was wondering if I could specify a custom metric when passing my data
frame into hclust as a distance matrix. Actually, kmeans doesn't even
take a distance matrix; it takes the data frame directly. I was
wondering if there's a way or if there's a package that lets you
create distance matrices from non-standard metrics, e.g.,
KL-divergence (which is not really one), but metrics used often in
information theory. I'm assuming kmeans just assumes the euclidean
metric, but it would be nice to customize that as well. Is stuff out
there, or would I have create my own?

Thanks,
Vivek

Bump...no insights on defining custom metrics. Guess I'll give the
other languages a shot.

Vivek

On Wed, May 5, 2010 at 10:13 AM, Vivek Ayer <vivek.ayer at gmail.com>
wrote:
> Hi guys,
>
> I've been using the kmeans and hclust functions for some time now and
> was wondering if I could specify a custom metric when passing my data
> frame into hclust as a distance matrix. Actually, kmeans doesn't even
> take a distance matrix; it takes the data frame directly. I was
> wondering if there's a way or if there's a package that lets you
> create distance matrices from non-standard metrics, e.g.,
> KL-divergence (which is not really one), but metrics used often in
> information theory. I'm assuming kmeans just assumes the euclidean
> metric, but it would be nice to customize that as well. Is stuff out
> there, or would I have create my own?
>
> Thanks,
> Vivek
>

---------- Forwarded message ----------
From: Vivek Ayer <vivek.ayer at gmail.com>
Date: Fri, May 7, 2010 at 10:17 AM
Subject: Re: [R] custom metric for dist for use with hclust/kmeans
To: Greg Snow <Greg.Snow at imail.org>


Hi Greg,

The pam function is exactly what I needed. I can now create my own
distance matrix, run as.dist on it and pass it into pam.

Thanks a lot!
Vivek

On Thu, May 6, 2010 at 12:42 PM, Greg Snow <Greg.Snow at imail.org>
wrote:
> The pam function in the cluster package accepts either raw data or a
dissimilarity matrix and does the same idea as kmeans. ?The daisy function has
more options for creating the dissimilarity matrix, if what you want is not in
there, you could still use it as a model for creating your own function. ?You
could also use the outer function and as.dist to create the distance matrix.
>
> --
> Gregory (Greg) L. Snow Ph.D.
> Statistical Data Center
> Intermountain Healthcare
> greg.snow at imail.org
> 801.408.8111
>
>
>> -----Original Message-----
>> From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-
>> project.org] On Behalf Of Vivek Ayer
>> Sent: Thursday, May 06, 2010 9:56 AM
>> To: r-help at r-project.org
>> Subject: Re: [R] custom metric for dist for use with hclust/kmeans
>>
>> Bump...no insights on defining custom metrics. Guess I'll give the
>> other languages a shot.
>>
>> Vivek
>>
>> On Wed, May 5, 2010 at 10:13 AM, Vivek Ayer <vivek.ayer at
gmail.com>
>> wrote:
>> > Hi guys,
>> >
>> > I've been using the kmeans and hclust functions for some time
now and
>> > was wondering if I could specify a custom metric when passing my
data
>> > frame into hclust as a distance matrix. Actually, kmeans
doesn't even
>> > take a distance matrix; it takes the data frame directly. I was
>> > wondering if there's a way or if there's a package that
lets you
>> > create distance matrices from non-standard metrics, e.g.,
>> > KL-divergence (which is not really one), but metrics used often in
>> > information theory. I'm assuming kmeans just assumes the
euclidean
>> > metric, but it would be nice to customize that as well. Is stuff
out
>> > there, or would I have create my own?
>> >
>> > Thanks,
>> > Vivek
>> >
>>
>> ______________________________________________
>> R-help at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-help
>> PLEASE do read the posting guide http://www.R-project.org/posting-
>> guide.html
>> and provide commented, minimal, self-contained, reproducible code.
>