13:20 < dk97[m]> zoq: I have added the loss functions folder along with the KL Divergence loss.

13:21 < dk97[m]> MSE and Cross-entropy are already implemented in the ann_layers folder, should I shift them to the loss_functions folder?

13:21 < dk97[m]> Also, I found some errors in the MSE implementation.

13:29 < apoorv_> Hi i am new to mlpack and follpw up the tutorials i want to contribute to the reinforcement leaening project can anyone give w the suggeations what is the most effective way to do that project..

13:35 < sshkhrnwbie> @zoq I finished up with the variance scaling initializer PR #1241 (from my end, yet to merge). I was going to get started on the randomized leaky relu as we discussed on 14/2

13:36 < sshkhrnwbie> @zoq I am going to implement it similar to dropout with a deterministic flag for randomising the activations during train time and fixing them during test. Does this sound right.

13:37 < sshkhrnwbie> @zoq Here is the link to the paper on arxiv for your reference https://arxiv.org/pdf/1505.00853.pdf

19:39 < luffy1996> zoq: Hey I am pretty free from hectic examination schedule. I would like to know if you went through my PR request. I would like to know whether I should proceed with implementing policy gradient methods on mlpack .

20:56 < daivik> rcurtin: Please forgive me if this is a stupid question, but is there a reason why there are two separate template params for each VecType in mlpack/core/metrics/lmetric.hpp:Evaluate() function? I mean, wouldn't we only want the distance between two arma::vecs or two arma::sp_vecs. I dont see where we would need to find the distance between an arma

20:56 < daivik> ::vec and an arma::sp_vec.. Moreover, attempting to do that results in a segfault (as it should). Isn't VecTypeB redundant?

20:57 < rcurtin> daivik: don't forget armadillo subview types; that's the real reason

20:58 < rcurtin> finding the distance between an arma::vec and arma::sp_vec should not cause a segfault, by the way

21:03 < daivik> https://thepasteb.in/p/lOhOKBYZ379SB --> this causes segfault for me - I assume this is an issue then?

21:07 < daivik> rcurtin: Also about the subview types -- do you mean something like distance between arma::vec and (some arma::mat).col(0) ?

21:09 < rcurtin> the code looks fine to me, it is strange that it causes a segfault. I looked in Evaluate(), I see little reason why what you have there shouldn't work, unless the vec and the sp_vec are not the same dimensions

21:09 < rcurtin> and yes, for the subview types, mat.col() returns a different type

21:11 < daivik> In the code, both the vec and the sp_vec are the same vector (and hence have same dimensions) - this can be seen by adding cout statements.

21:45 < daivik> rcurtin: I'll look into this a bit more tomorrow morning and let you know what I find.. Thanks for your help :)