On 18/6/09 9:52 AM, Stephanie Januchowski wrote:> Hello List,
>
> My question is an elementary one. I have run a fuzzy kmeans cluster using
> FANNY to group freshwater fish assemblages. I then went in the field to
> validate that classification and have retrieved new assemblage data for a
> new suite of streams. Therefore I would like to use Predict to determine
how
> well the original clustering fits the new data. However I have not figured
> out a straightforward way to get predict to work with the outputs generated
> from FANNY. I have generated a dissimilarity matrix with vegdist and
figured
> I could predict the classification onto the new data using that matrix,
> however R keeps telling me there is no predict application for the code I
> have written.
>
> Any suggestions would be more than appreciated.
>
> Thank you all for your time.
>
> Stephanie
>
It sounds like you're really more interested in (supervised)
classification rather than (unsupervised) clustering. If there are
natural classes for the data, e.g. from literature, just use a
classifier like logistic regression or SVMs, train on your first group
of data and test on the new data. If you don't know which classes are
possible, and you want to find possible classes in the data, you can use
the clustering to identify important variables and find the classes,
then use these variables to classify the fish into classes (using a
classifier) and then check whether the other data agrees with your
classification.
Hope it helps.
--
Gad Abraham
MEng Student, Dept. CSSE and NICTA
The University of Melbourne
Parkville 3010, Victoria, Australia
email: gabraham at csse.unimelb.edu.au
web: http://www.csse.unimelb.edu.au/~gabraham