[R] KNN script: Identity of specific K samples chosen?

[Max Kuhn]

> I am currently doing some prediction work using the knn script in the
> 'class' package. Does anyone know a way of having R return the IDs (sample
> IDs, or column IDs of the training matrix) of the 'k' samples that are
> chosen by the algorithm as being nearest to a given test sample?

You would probably have to modify the source code to get the specific neighbors.

> I have searched/read everything I can about the script, however have not
> found anything other than the ability to output the proportion of 'k'
> samples agreeing with the final prediction (which is also quite useful).

The knn3 function in the caret package will return the votes per class
for each test sample.

> However I would still like the 'k' sample ID's, for further lookup and
> reporting steps in my process.

Since this sounds like a research project, a better idea might be to
use the proxy package (or some other) to compute the neighbors
yourself and do what you wish with them.


Max