03:24 < pvskand> @rcurtin I was planning to start RBM module. I had mailed marcus about it but I guess he is busy. Should I first create an issue and then start off with the module?

03:25 < rcurtin> pvskand: I'm sure you can see, there is a huge amount to process for GSoC, so everyone is very busy

03:25 < rcurtin> is there an issue already open for RBM support? or, does anyone know that you will be contributing RBMs?

03:25 < rcurtin> in either case, you should post what your design will look like, so that it can be reviewed before you implement it

03:26 < rcurtin> and keep in mind, it may be some time before anyone is able to review your design, so please be patient... we receive very many emails and requests every day :)

04:18 < vivekp> zoq: rcurtin: Regarding SMORMS3 policy class; I'm trying to resolve a very peculiar test failure. So far, I've isolated the bug using valgrind

04:19 < vivekp> Looks like in iterate update step, I can't do the Schur's product of gradient and what I've called temp.

04:19 < vivekp> Here's what SMORMS3Update class looks like: https://paste.debian.net/919967/

04:19 < vivekp> and valgrind reports this error: https://paste.debian.net/919966/

04:19 < vivekp> Any suggestions?

04:23 < vivekp> Valgrind doesn't complain so long as I'm doing iterate -= gradient; but really we want iterate -= gradient % temp;

04:52 < pvskand> @rcurtin yes I do understand that you are really busy! I will be waiting in the mean time.

08:50 < kamal_____> hey! i a new here and i am willing to contribute! please help me get started. i am comfortable with c++

14:51 < travis-ci> mlpack/mlpack#2033 (master - 957f446 : Ryan Curtin): The build was fixed.

14:51 < travis-ci> Change view : https://github.com/mlpack/mlpack/compare/3b9917fad988...957f4461cc9b

14:51 < travis-ci> Build details : https://travis-ci.org/mlpack/mlpack/builds/211363364

14:53 < rcurtin> vivekp: are you sure that 'gradient' and 'temp' have the same size? the valgrind output seems to me like maybe the schur product is going past the end of one of the arrays

14:53 < rcurtin> one thing you can do to check, is compile in debugging mode; this way, Armadillo will check sizes before operations, and you will get a more clear runtime error if the sizes aren't right

14:53 < rcurtin> you can reconfigure CMake with the '-DDEBUG=ON' option and then rebuild and re-run

14:54 < rcurtin> when you do that valgrind will give you slightly nicer output too, since it will be able to print the line number of any problems :)

15:07 < vivekp> thanks a lot! I'll try that and report back.

15:18 < rcurtin> yeah, let me know what you find. valgrind is a great tool for hunting stuff like this down

16:02 < vivekp> rcurtin: yeah, I agree. Finally fixed the errors.

16:03 < vivekp> You rightly pointed out the size issue -- I had somehow messed up the matrix initializations in the Initialize method.

16:03 < vivekp> Thanks again :)

17:30 < kartik__> <zoq> I have done the cmaes super mario code also .. fixed issues and started running it ..

17:30 < kartik__> would request you to once to go through the supermario.cpp file

17:30 < kartik__> here it is

17:30 < kartik__> https://github.com/kartik-nighania/CMA-ES_SuperMario/tree/master/SuperMarioBros

17:32 < kartik__> also i have implemented CMAES with FFN modified by Bang genome .. plz take a look on PR #938

17:32 < kartik__> Both test have passed ..

17:35 < rcurtin> kartik_: kartik__: I'm sure he's already seen your PR, I think everyone subscribed to the list gets an email every time a PR is opened, so don't worry, it will get looked at when there is time

17:36 < kartik__> <rcurtin> thanks. He can have a look whenever he wants..

17:37 < kartik__> <rcurtin> this is my first open source org.. i have never written test before.. can u suggest me some good links ?

17:42 < zoq> kartik__: Should I look for something specific in supermario.cpp?

17:42 < zoq> kartik__: Also with tests passed I think you mean some of the easier tasks like CartPole?

17:43 < kartik__> yes ..actually the mario isnt working correctly .. needs some sort of parameter tuning..

17:44 < kartik__> your look at cpp would help

17:45 < zoq> kartik__: hm, so it kinda works, but you can't see any progress?

17:46 < rcurtin> kartik__: sure, I am just saying, there is only need to notify him once, I see like three or four different notifications through IRC and email

17:47 < kartik__> yes .. saw him moving in just one direction.. also i think the initial std deviation and 0.5 inital weights for all FNN is causing problem maybe.. will work on it overnight

17:50 < kartik__> <rcurtin>oh , sorry for that .. i will not make sure of it doesnt happen in future..

17:50 < kartik__> <rcurtin> i mean .. i will make sure ..

18:29 < gautam> Hello Everyone, Im an undergraduate student from Mumbai University.I have experience with python,web development and recently started studying machine learning from Coursera and Deep Learning from Udacity using Tensor Flow.I am interested in contributing to the Low rank/sparse optimization using Frank-Wolfe problem and have downloaded and installed mlpack, run a few tutorials. I am good with algorithms and my math knowledge is strong

18:29 < gautam> Wanted to know whether I would be able to contribute for GSoC'17 with the given problem.I have very basic knowledge of C++.This is my first foray into open source contribution..

18:30 < gautam> Thanks , Gautam Worah.

19:27 < kartik__> <gautam> u can go through http://www.mlpack.org/involved.html to start contributing ..

19:28 < kartik__> also https://github.com/mlpack/mlpack/wiki/SummerOfCodeIdeas is the page for ideas. you can start working on them or come up with your own proposal :)

22:36 < s1998> :zoq :rcurtin

22:36 < s1998> I am facing a new issue

22:36 < s1998> CLI_test is failing

22:36 < s1998> Error says "[FATAL] Required option --input_file is undefined."

22:37 < s1998> I tried to isolate the issue and found that it is coming from line "CLI::ParseCommandLine(argc, argv);"

22:38 < s1998> I tried to search on net but wasn't able to find anything!! Any help please ?

22:40 < s1998> I wasn't able to find definition of this function too !!

22:40 < rcurtin> s1998: can you tell me what you are doing to cause the issue?

22:40 < rcurtin> and if you have made any modifications to the code?

22:41 < s1998> I was just running the tests and this happened

22:41 < rcurtin> I am about to get dinner, so I can't stay for more than a few more seconds

22:41 < s1998> I only added HDBSCAN method

22:41 < s1998> Oh !!

22:41 < rcurtin> in the HDBSCAN tests are you doing anything with CLI?

22:42 < s1998> Yes I am trying to get input and ouputs using this !!

22:42 < s1998> But then cli_test should not faill, ryt ?

22:42 < s1998> *fail

22:56 < s1998> I am trying to use CLI only in hdbscan_main, not in hdbscan_tests

23:37 < kris1> zoq: Did you have chance to go through this https://gist.github.com/kris-singh/e1b4866273c2e809ad1deef398ce5103

23:39 < kris1> This is not training well. i am not able to figure it out. Do you see any obvious mistakes in that

23:43 < zoq> kris1: Unfortunately, I haven't had time to look into it, but it's on my list.

23:45 < kris1> no worries. please let me know if you do.

23:45 < zoq> kris1: I'll keep you updated.

23:46 < kris1> Thanks

23:47 < kris1> zoq: for deep learning project. Are you guys also interested in variational autoencoders,

23:48 < kris1> I have read about them. Have worked a little with them when i contributed pgmpy and pymc3.

23:50 < zoq> The project idea is pretty open, so if you find something that looks really interesting and you think you can implement a fast version of the idea, I'm open for discussions.

23:52 < zoq> Just glanced over the code, when you do boost::apply_visitor(ParametersVisitor(std::move(weights)), model.Model()[model.Model().size() - 1]); have you checked that the parameters are actually updated e.g. by looking at model.Parameters()?

23:53 < kris1> zoq: Yes i have. Actually i was following the karpathy is code for implementation.

23:54 < kris1> They used updated the weights += gradient * step

23:54 < kris1> i did not understand that part.

23:58 < zoq> That is exactly what we do in the update process if we use standard sgd. If you checked the parameters, I'll have to dig into some details of the code ... as soon as I get the chance