[mlpack-git] [mlpack] Different matrix type for kmeans::RefinedStart (#378)

[Ryan Curtin]

This is related to #290 (that ticket references #300 but should reference #288; the incorrect reference is due to a semi-failed import from Trac...).  Basically, mlpack is currently hard-coded to work with doubles, but templatizing `MatType` for every class could allow arbitrary precisions and types (within the confines of what Armadillo provides).  That would be a ton of work, though, and the additional testing required would also be difficult to provide...

For now your best bet is probably to cast your data to `arma::mat`, then cast your resulting centroids back to `arma::Mat<uint8_t>`.

As for whether or not the clustering makes sense, yeah, it makes sense.  Even if the clustering was done in the `uint8_t` space I think it would converge fine.

---
Reply to this email directly or view it on GitHub:
https://github.com/mlpack/mlpack/issues/378#issuecomment-69590883
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.cc.gatech.edu/pipermail/mlpack-git/attachments/20150112/7ccd0cac/attachment.html>