04:36 < Shubhankar> Hlw

07:36 < Soonmok> Hi! I send a pull request to mlpack/models. I added gan application on the Mnist dataset and generate images.Please give some feedback if you have spare time. Thanks

07:49 < travis-ci> Soonmok/models#1 (master - b30607c : Soonmok): The build failed.

07:49 < travis-ci> Change view : https://github.com/Soonmok/models/compare/d81c59e2b019...b30607c583db

07:49 < travis-ci> Build details : https://travis-ci.com/Soonmok/models/builds/103155075

08:03 < travis-ci> Soonmok/models#2 (master - 12eb8ca : Soonmok): The build failed.

08:03 < travis-ci> Change view : https://github.com/Soonmok/models/compare/b30607c583db...12eb8ca0bd24

08:03 < travis-ci> Build details : https://travis-ci.com/Soonmok/models/builds/103156165

08:12 < Alex_> I have a query regarding GSoC. I have narrowed down a project from the given ideas list which I want to work on. I have already started studying relevant papers for the same. What factors will mlpack take into consideration if multiple students apply for the same project idea.Thanks

10:26 < ShikharJ> Alex_: We base our decisions on the quality of the proposal, the student's experience and prior ability to get things implemented in mlpack, and most importantly, the availability and interest of the concerned mentors :)

10:27 < ShikharJ> Soonmok: Thanks for the PR. I'll take a look soon.

12:27 < Prashant> join

12:31 < Kd_> Hi how i can run those test for command-line and Python bindings

12:33 < Kd_> Let's say I want to run kde_test.cpp than I can run it directly by gcc kde_test.cpp because it given lots of error but no testing page.

12:35 < Kd_> Can anybody tell how i can run those tests .

13:22 < rcurtin> Kd_: use CMake to make 'mlpack_test' then run, e.g., 'mlpack_test -t KDEMainTest'

13:26 < rajiv_> I have implemented dense blocks class but I'm not sure about it's correctness... What test should I use? I had referred the Keras implementation to make it... So can I use it as a benchmark or do you suggest any other testing method?

13:55 < robb_> I'm using an RNN. My data is a cube of size(1,5,2), and my labels are in a cube of size(1,1,2). Why could I getting a slice() out of bounds error?

13:56 < robb_> be getting*

14:43 < mulx10> robb_: Hello

14:43 < mulx10> robb_: I guess you are forgetting the `rho` param

14:45 < mulx10> try keeping a cube of size(2,5,2) and a cube of size(2,1,2) for data and labels repectively.

14:45 < mulx10> I guess this should resolve the error

14:46 < mulx10> robb_: Do let me know if it works.

15:36 < robb_> mulx10: That didn't work either, my rho is 5... Not sure if I'm conceptualizing it wrong

15:36 < robb_> here is the short test code https://github.com/RMaron/mlpack-rnn-test/blob/master/main.cpp

15:37 < robb_> it's just supposed to predict the next number in the +1 sequence

15:45 < mulx10> robb_ : then use cube of size(5,5,2) and a cube of size(5,1,2)

15:47 < mulx10> robb_: the point is the data must be expanded to incorporated the rho param

16:27 < mulx10> robb_: this (https://pastebin.com/KZpyCzRs) might help.

17:30 < Gilles_> Hey everyone, I'm a Belgian undergraduate student in Engineering looking to participate in GSoC 2019. mlpack is one of the organisations that caught my eye, though I don't really know much about machine learning yet (I'm more than willing to delve into it though, as it's what I want to pursue after my undergrad degree is finished). I had a look at the proposed projects, and found the 'Improvement of tree traversers' one to look the

17:32 < Gilles_> I'm quite new to C++, but have some experience with other programming languages and assume it won't be too hard to learn C++ up to a decent proficiency. What are some ways for me to get started on learning how to tackle this project, other than the basic "build mlpack and run the examples" advice?

18:29 < rcurtin> Gilles_: hi there

18:29 < rcurtin> I think your first message got cut off (you can see how much of it we saw on http://www.mlpack.org/irc/)

18:30 < rcurtin> the tree traverser improvement project is a pretty difficult project, so as you dig into it, do feel free to maybe branch out and consider some different projects

18:31 < rcurtin> in any case, I would suggest for that one that there's a bug open right now, #1269, that points out a problem with spill trees that has to do with the tree traversers

18:31 < rcurtin> there are already some PRs open for this, but I haven't had a chance to review them

18:31 < rcurtin> so, perhaps a thing to do might be to look at the "tree-independent dual-tree algorithms" paper to understand a bit more about the computations being done and what the traversers are doing,

18:32 < rcurtin> and then you could take a look at the PRs that are open (you can leave reviews, too, it's perfectly okay), and see if they are wrong or right or what

18:32 < rcurtin> I need to review those but I haven't had a chance yet...

18:52 < saksham189> zoq: ShikharJ I had a few queries regarding the Essential Deep Learning Modules idea for gsoc.

18:54 < saksham189> can i propose my own choice of modules that I wish to implement or are there any modules that would be high priority for mlpack?

18:57 < saksham189> I was thinking about implementing GAN variants and Deep belief network.

18:58 < rcurtin> saksham189: I'll let either of them correct me if I'm wrong, but I'm relatively sure the answer will be that you can definitely propose your own choice of modules

18:59 < rcurtin> I can't say if there are any that are currently high priority though. Sumedh's Neural Turing Machines project and Konstantin's Highway Networks were really cool, but they never ended up getting merged, since there was still a little work left to do

18:59 < ShikharJ> saksham189: Feel free to propose your own choices, as Ryan said.

19:00 < ShikharJ> I agree with Ryan, I think we have even mentioned how getting the pending PRs merged in is the highest priority work.

19:00 < rcurtin> it's definitely a nice "warmup" to get familiar with the codebase, to review the unmerged PRs, fix what's wrong, and get them "across the finish line" and merged :)

19:01 < rcurtin> the only issue is finding the time to review the updated PRs... I am having a lot of trouble with that right now...

19:06 < saksham189> Cool! I will take a look at the highway networks PR and see if I can complete that.

19:18 < rcurtin> sounds good; I don't remember the state that it was in, I think it needed a bit of work...

19:31 < Gilles_> eh, the part that got cut off was just the end of the sentence, it's not even worth retyping. I'll have a look at #1269 and will try to find the time to read the paper you mentioned, too. is there any required knowledge to be able to understand it, other than the basics of trees in graph theory?

20:29 < rcurtin> Gilles_: it's definitely helpful to have some understanding of what a kd-tree is, but not entirely necessary

20:29 < rcurtin> the paper explains that, but also it can sometimes help to have an intuitive grasp of a kd-tree so the definitions make sense

20:30 < rcurtin> you can also read my thesis (http://ratml.org/pub/pdf/2015improving.pdf) but... it's long and not guaranteed to be interesting

20:30 < rcurtin> there is a picture of my cats hidden in there somewhere though... :)

20:31 < travis-ci> mlpack/ensmallen#171 (master - 64f15cf : Marcus Edel): The build has errored.

20:31 < travis-ci> Change view : https://github.com/mlpack/ensmallen/compare/964b6801ac94...64f15cf9be21

20:31 < travis-ci> Build details : https://travis-ci.org/mlpack/ensmallen/builds/502225503

21:10 < Gilles_> rcurtin: i'm getting the impression that you really don't care about formalities, and really love your cats...

21:10 < Gilles_> neither of which bother me in the slightest, that being said.

21:18 < rcurtin> :)

21:18 < rcurtin> it could be true :)