[mlpack] Regarding RBMs

[Udit Saxena]

Hi,

On Tue, Feb 3, 2015 at 1:53 AM, Ryan Curtin <ryan at ratml.org> wrote:

> On Sun, Feb 01, 2015 at 08:28:45PM +0530, Udit Saxena wrote:
> > Hi Marcus,
> >
> > I've put up a basic CD-1 RBM up on:
> >
> >
> https://github.com/uditsaxena/mlpack-1/tree/master/src/mlpack/methods/rbm
> > <
> http://s.bl-1.com/h/mkx4S4z?url=https://github.com/uditsaxena/mlpack-1/tree/master/src/mlpack/methods/rbm
> >
> >
> > I think you can have a look at it now.
> >
> > Also, I'm getting stuck on a "does not name a type" error, which I feel
> is
> > because I've messed up somewhere during header files declaration. I was
> > wondering if you could have a look at the error as well.
>
> Looks like you've solved the issue with some later commits; or are you
> still having trouble?  If so I'll dig in a bit deeper.


I haven't, unfortunately. I tried, but couldn't get it. Though I'm pretty
sure it's just a library linking error. Nevertheless, please have a look.

>
> > I was wondering about the kind of tests that could be used for the RBM.
> Any
> > thoughts ?
>
> One idea might be to find some papers that use RBMs very similarly to
> how you have implemented, and ensure that the training and test errors
> are within some margin of what the paper says RBMs are able to do.  This
> is how Sumedh and I did some of the SVD tests.


Okay. I was also thinking of using some theorems/corollaries we could end
up verifying, the same way we did for adaboost, perceptron and decision
stumps.

Thanks !

>
> --
> Ryan Curtin    | "And they say there is no fate, but there is: it's
> ryan at ratml.org | what you create." - Minister
>



-- 
---------------------------------------
Udit Saxena
Student, BITS Pilani,
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cc.gatech.edu/pipermail/mlpack/attachments/20150203/1972c6a2/attachment-0003.html>