Commons math typically depends on very specific data representations which are not very compatible with Mahout's needs.
We have used commons math stuff in the past, but it requires a copy operation. That often kills any possible benefit. The clustering algorithms are similarly conflicted and there, there are also problems with scaling. On Thu, Jun 12, 2014 at 5:06 PM, Ravi Mummulla <[email protected]> wrote: > Is someone looking at refactoring several of the Mahout algorithms to take > a dependency on commons.math where there are overlaps? For example, matrix > decomposition from the commons.math linear algebra library, some of the > clustering algorithms (k-means, fuzzy k-means, etc.) from commons.math > machine learning library. > > Thanks. >
