00:33 < kris1> The error i think is that there isn’t enough memory

01:08 < rcurtin> blah, there goes gitdub again sending tons of commit messages it already sent

01:45 < lamacarena> Hey. is mlpack comparable to tensor flow?

02:11 < travis-ci> mlpack/mlpack#2876 (mlpack-2.2.4 - ae8e35b : Ryan Curtin): The build passed.

02:11 < travis-ci> Change view : https://github.com/mlpack/mlpack/compare/mlpack-2.2.4

02:11 < travis-ci> Build details : https://travis-ci.org/mlpack/mlpack/builds/255086664

07:07 < lozhnikov> kris1: I think there is no quick answer since I should dig into the code and debug that a little in order to understand the issue. I'll look through the code today.

07:08 < kris1> Okay…just to point out the code works till size of inSize = 64 and outSize = 100 so you can try with something like 100, 100

09:42 < partobs-mdp> zoq: What should I do to wrap up the task PR aside from removing AddTask?

09:44 < partobs-mdp> zoq: As a follow-up: what models shall we use for creating first working implementation of HAMUnit?

09:45 < partobs-mdp> zoq: As a first approximation, I offer to use linear functions for Join/Search/... modules of HAM. What do you think?

09:55 < zoq> partobs-mdp: I'll take a look at the PR and make some comments, later today.

09:55 < zoq> partobs-mdp: Also, starting with a linear function sounds good to me.

09:57 < partobs-mdp> zoq: Ok, so I propose this: affine functions for EMBED/JOIN/WRITE, sigma(affine function) for SEARCH.

09:58 < partobs-mdp> zoq: Now about gradients: we have ∇log p(y | A, x, θ) and ∇log p(A | x, θ)

09:59 < partobs-mdp> The latter one seems to be the gradient of SEARCH function

10:00 < partobs-mdp> The first one... well, I don't really understand the first one - can you explain to me what is that?

10:00 < partobs-mdp> *The latter one seems to be the gradient of SEARCH function wrt SEARCH function weights (btw, am I right here?)

10:03 < zoq> partobs-mdp: I'm in a meeting, I get back to you once I have a chance, might take some hours :(

15:29 < lozhnikov> kris1: It seems the formulas in the paper are correct except the sign near s_i^T \alpha_i s_i in the energy function. This minor issue doesn't affect sampling formulas but affects free energy. Here is the fix. https://usercontent.irccloud-cdn.com/file/zUKoAjS2/energy.jpg

15:31 < kris1> Okay i will have a look thanks.

15:33 < kris1> I will open the gan pr later today.

15:47 < kris1> Also were you able to test the memory error i was taking about

16:45 < kartik_> <zoq> i think zoq CMAES cant be applied to logistic regression as u said because there is no error calculation like SGD and using the gradient to minimize. The build of logistic regression just passed but it doesnt leads to any output

18:22 < travis-ci> mlpack/mlpack#2878 (master - 923df37 : Ryan Curtin): The build was broken.

18:22 < travis-ci> Change view : https://github.com/mlpack/mlpack/compare/251a6a9a5f42...923df37f9e36

18:22 < travis-ci> Build details : https://travis-ci.org/mlpack/mlpack/builds/255356977

19:59 < MikeLDN> Hi. I need a hint... stucked with LNK 2005 error in VC2015 (and 2.2.4)

20:00 < MikeLDN> I have a working CF demo dialog app and when I add #include <mlpack/methods/range_search/rs_model.hpp> bellow other #include <mlpack/methods... I get LNK errors with visitors

20:00 < MikeLDN> "public: void __cdecl mlpack::range::TrainVisitor::operator()(class mlpack::range::RangeSearch<... already defined, in both app.obj and appdlg.obj

20:12 < rcurtin> MikeLDN: I'm not too familiar with visual studio, so I don't know how much help I can be, but my best guess is that what's happening is that you're instantiating a CF object in both app.cpp and appdlg.cpp

20:12 < rcurtin> this causes the same symbols to be instantiated twice, which I guess causes a crash with the VS compiler

20:13 < rcurtin> the only thing I can think, and it's kinda ugly, is to write a "wrapper class" around the CF class, so all the mlpack .obj symbols for CF will be contained in that

20:13 < rcurtin> maybe there is some other way to do it in VS, like I said I am not too familiar with it

20:25 < MikeLDN> Thnx. It is actually related to the range search. I wasn't able to make a RS test app so I just "dropped" include rs_model.hpp into the working CF test app, to check it...

20:26 < MikeLDN> VC gives the same error. Anyway, command line mlpack_range_search is compiling and working OK so I will try to compare with my project and see where the difference is...

20:27 < rcurtin> it might have to do with some linker configuration; it looks like the compiler is generating all the RangeSearch symbols twice when it only needs to do it once