OK.

I have a first pass at converting the colt eigen solver to our math
package.  It wasn't hard with the help of a few emacs macros, but in the
process I found what really look like some bugs.  That throws the utility
of the conversion into question.  At the same time I wrote a from-scratch
version that is based on using SVD (only works on symmetric matrices).  I
also started in on an idiomatic version of the Colt decomposer that users
our vectorized operations.

Next would be to build some test cases and figure out what is happening.  I
may have to fall back to the JAMA or commons math libraries for inspiration
if the Colt versions turn out flawed.

Overall, this isn't any worse than I expected.

On Wed, Apr 25, 2012 at 2:12 PM, Dmitriy Lyubimov <[email protected]> wrote:

> if reliable eigensolver is available, we can also get rid of
> dependency on eigensolver from apache math in SSVD. Don't quite know
> if that's the only dependency on apache math, i saw one somewhere else
> as well.
>
> On Wed, Apr 25, 2012 at 8:09 AM, Ted Dunning <[email protected]>
> wrote:
> > I would like to take a stab at a collections/math merge and a deletion
> if deprecated math stuff. The deletion will require me to write and eigen
> solver to support some of Jake's old code.
> >
> > Sent from my iPhone
> >
> > On Apr 25, 2012, at 9:52 AM, Jeff Eastman <[email protected]>
> wrote:
> >
> >> Currently there are 24 outstanding JIRAs targeted for 0.7, which is
> intended to be a clean-up and housekeeping release. They are about evenly
> split between high and low priority tasks. Seven are marked asPatch
> Available. Now might be a good time to take a look at them to triage the
> list and begin to wrap up the remaining issues.
> >>
> >> Jeff
> >>
> >> PS: I've completed my move to Erie PA and should have quite a bit more
> time to devote to Mahout
>

Reply via email to