[mlpack] [mlpack/mlpack] adds GammaDistribution::Train(observations, probabilities) (#834)

[Ryan Curtin]

@yashu-seth: thanks for the contribution, sorry it took me a little while to get around to looking at this.  Don't worry about the test failure, that is for the Nystroem method, it doesn't have to do with your code.  Some of the mlpack tests are probabilistic and although we try to keep the failure probability very low, in some cases the probability isn't low enough.

The testing method seems reasonable---I think the gamma distribution that you'll fit with the uniform distribution for weights should have approximately the same parameters as if you trained it without weights.  Here are another couple ideas for simple tests:

 * Draw points from two different gamma distributions.  Set the probabilities for the points from the first distribution to something small (0.01, 0.001, something like that) and the probabilities for the second to something large (0.99, 0.999, something like that).  The gamma distribution that you recover should have the same parameters as the second gamma distribution you drew from.

 * Train with probabilities all set to 1, and ensure that this gives the same result as when you train with no probabilities at all.

Hope this helps---let me know if I can clarify anything.  I glanced at the implementation, I think it looks right; just needs tests. :)

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/mlpack/mlpack/pull/834#issuecomment-268355068
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://knife.lugatgt.org/pipermail/mlpack/attachments/20161220/965c2ba9/attachment.html>