06:01 < pegasus> hey, amond the ann layers of mlpack, the merge modes available are for add and concat.

06:03 < pegasus> although they are the more commonly used ones, libraries usually have average, multiply etc. modes for merge

06:03 < pegasus> Will implementing them be a good addition for mlpack?

12:31 < rohit> hi i just installed ml-pack on my laptop. Can someone tell me what does /core directory contains ?

13:06 < zoq> Classes that are used by other methods often by more than one methods, like optimizer that are used for logistic regression, neural networks etc.

13:07 < zoq> Or classes to load/save matrices.

13:08 < rohit> do i need to have their knowledge in order to contribute ?

13:36 < zoq> no, in most cases you can just use the classes without looking into the code.

14:49 < rohit> oh thanks :)

16:00 < alsc> zoq : so I am now eventually coding the TerminationPolicy

16:00 < alsc> besides the DecomposableFunctionType, I'd like to have the last objective as a parameter how does that sound?

16:02 < alsc> because ok if I pass the model I can test the validation accuracy, but I'll really want to have both training and validation loss

16:13 < alsc> rcurtin: there?

16:15 < alsc> I guess I'll just make members some auto variables in SGD::optimize, the full state will be available to the caller-back then

16:53 < rohit> its strange,the mlpack i got on my laptop has no directory as src/mlpack/bindings ?

17:14 < alsc> so ok I have this down... applies to rmsprop, ada_delta, ada_grad and rmsprop

17:14 < alsc> https://github.com/mogees/mlpack/commit/4adbc90454c8ec2057b9dcbd32ed29c788815423

17:16 < alsc> could you guys review this and we work towrds a PR? it doesnt break the existing code... only thing in order to get this flexibility one should make template rmsprop, ada_delta and ada_grad... me myself, I'll prefer to use SGD directly without these wrapper classes that are used just to set the correct UpdatePolicy... let me know