00:03 < travis-ci> mlpack/mlpack#4084 (master - f465faf : Marcus Edel): The build passed.

00:03 < travis-ci> Change view : https://github.com/mlpack/mlpack/compare/a615ba74399a...f465faf9d81a

00:03 < travis-ci> Build details : https://travis-ci.org/mlpack/mlpack/builds/345438479

00:05 < rcurtin> ok, so it turns out that in order to get the ports opened, we need approval from another department

00:05 < rcurtin> they have expedited the request but will have it done Monday at the earliest

00:06 < rcurtin> I suppose it's possible I could tunnel port 80 on masterblaster through mlpack.org

00:07 < zoq> "Monday at the earliest" ... hopefully

00:18 < rcurtin> I got tired of waiting

00:18 < rcurtin> http://mlpack.org:8180/

00:18 < rcurtin> not sure how long that connection will last, but I can probably get some PRs to build after updating the hooks

00:25 < rcurtin> actually it also seems like this will not work fully correctly because the URLs are not properly rewritten, so maybe it is easier to just wait for Monday :-\

04:59 < luffy1996> Every one my test errors were build with out successfully. However testing my code with mlpack_test shows SIGBART error. Anyone having any intuition why it is caused?

05:00 < luffy1996> https://www.irccloud.com/pastebin/GLD7TWmY/

05:01 < luffy1996> I have not posted the entire log. This is the last three lines of the log

05:14 < rcurtin> luffy1996: try using gdb to trace what the error is

05:15 < luffy1996> Could you please more elaborate :)

08:06 < moksh> @zoq I wanted to clarify something, I should implement a model(say policy gradients) for one of the available environments(like cartpole) or implement some other environment support?

10:15 < nilkel> Hi everyone, Neel here. I'm interested in participating in GSOC 2018 as a student. I've built mlpack from source and I've gone through the tutorials and documentation. I was hoping to work on implementing the deep learning modules and I'm going through the literature right now. I'm still a noob to mlpack though so could someone point me in the right direction and help me get started?

11:57 < _Mail> Hey people

11:58 < _Mail> I got to get started with ML Pack

11:58 < _Mail> Need help to start

11:58 < zoq> _Mail: Hello have you seen: mlpack.org/gsoc.html and www.mlpack.org/involved.html?

11:59 < _Mail> Thanks zoq

11:59 < zoq> luffy1996: As rcurtin pointed out take a debugger and see if you can track the error down.

12:00 < zoq> _Mail: Here to help :)

12:01 < zoq> moksh: Hello, that is up to you, you could implement a simple RL method or implement another test environment.

12:04 < zoq> nilkel: Thanks for getting in touch, if you search for general advice please take a look at mlpack.org/gsoc.html and www.mlpack.org/involved.html. There are some open discussions on the mailing list which could be helpful.

13:51 < dk97[m]> hey there

13:52 < dk97[m]> I saw some activation functions in the ann module

13:58 < zoq> dk97[m]: Hello there, yes.

13:59 < dk97[m]> However elu, celu activation are not implemented there, rather they are implemented in the layers

13:59 < dk97[m]> Any particular reason?

13:59 < dk97[m]> @zo

13:59 < dk97[m]> zoq:

14:02 < zoq> Yeah, the elu layer doesn't implement the same interface as the sigmoid function, since it holds another parameter (alpha). To set the additional parameter as construction time, we have to provide another constructor. I hope this makes sense.

14:05 < dk97[m]> so, point wise functions are implemented in the activation functions, but activation layers needing other layer parameters or parameters at construction time are implemented in the layers module.

14:05 < dk97[m]> zoq:

14:05 < zoq> Correct

14:09 < dk97[m]> https://arxiv.org/pdf/1706.02515.pdf

14:09 < dk97[m]> This paper has a new function selu

14:10 < dk97[m]> In this, alpha and lambda are fixed by definition.

14:10 < dk97[m]> So where should I include this function?

14:10 < dk97[m]> I am a bit confused...

14:10 < dk97[m]> zoq:

14:15 < zoq> dk97[m]: If both parameters are fixed, which they are let's implement the function as another activation function.

14:16 < dk97[m]> so should I work on this and submit a PR for the same?

14:16 < dk97[m]> zoq:

14:16 < zoq> If you like to yeah sure :)

14:17 < dk97[m]> cool then, I will let you know the progress!

14:17 < zoq> Sounds good.

14:22 < dk97[m]> do you use any particular code linter? I am using clang format 4.0

14:22 < dk97[m]> zoq:

14:23 < zoq> dk97[m]: See https://github.com/mlpack/mlpack/wiki/DesignGuidelines and https://github.com/mlpack/mlpack/pull/1205.

14:26 < dk97[m]> thanks I will have a look! zoq

14:49 < manthan> support for the use of word vectors or GLoVE can be added for string processing. Already available word vectors can be used for representing strings. Can this be done for "string processing utilities" idea for GSoC 2018?

15:46 < manish7294> rcurtin: I went through armadillo codebase and I found that armadillo uses " for(uword i=0; i<n_elem; ++i) {}" for calculating "

15:46 < manish7294> arma::uvec results = (mat == SIZE_MAX); "

15:46 < manish7294> . So, I think we may be able to use arma::accu() in place of loop :]

15:49 < manish7294> " n_nonzero += (A[i] == val) ? uword(1) : uword(0); " manages the calculation.

16:00 < caladrius> Hi! I am Aarush from India. I am a Deep Learning enthusiast and have worked on a couple of related projects. I am interested in contributing to mlpack under the GSoC program. I went through the introductory pages and am currently trying out some examples. I am also going through the ANN part of the codebase(as suggested in the chat archives). I am interested in working on a sequence to sequence encoder decoder network. Is it feasi

16:00 < caladrius> ^Also, no work has been done on Siamese Networks. Could I also work on that? Thanks!

16:02 < rcurtin> manish7294: sounds good, thank youbfor digging so deep :)

16:03 < rcurtin> caladrius: sure, maybe if you write a seq2seq example we can merge it into the mlpack/models repo

16:05 < caladrius> Okay, I will work on that. Could we do something for Siamese networks as well?

16:07 < manthan> rcurtin: can you please guide me regarding the use of word vectors or GLoVE for string processing utilities.

16:14 < manish7294> rcurtin: Looks like tests related to calculation of noise points have been removed from DBSCAN_Bindings_test PR. Seems like, I missed the disscusion :)

17:12 < Prabhat-IIT> Hi I couldn't find any documentation regarding ann modules in official mlpack 2.2.5 docs! Plz can anyone direct me to docs?

17:15 < zoq> Prabhat-IIT: Currently the best option is to take a look at the tests: https://github.com/mlpack/mlpack/blob/master/src/mlpack/tests/feedforward_network_test.cpp, https://github.com/mlpack/mlpack/blob/master/src/mlpack/tests/recurrent_network_test.cpp, https://github.com/mlpack/mlpack/blob/master/src/mlpack/tests/ann_layer_test.cpp

17:15 < zoq> Prabhat-IIT: A tutorial will come in the next days.

17:16 < zoq> caladrius: Yes we can do the same for Siamese networks.

17:32 < caladrius> @zoc Okay. Can a maxout layer be implemented? It would be useful for approximating functions

17:34 < caladrius> zoq:

17:34 < zoq> caladrius: No, but that is straightforward to implement, let me know if you need any help.

17:35 < caladrius> okay thanks!

17:36 < caladrius> zoq: So there's no need to implement it, right?

17:36 < zoq> caladrius: Maxout?

17:36 < caladrius> yeah

17:37 < zoq> I think we could just use arma::max over the output, but you could implement it as an extra layer.

17:37 < zoq> Both should work just fine.

17:38 < caladrius> I'll try it out then. Thanks!

17:41 < zoq> Sounds good :)

17:43 < kaushik_> zoq: do we have a tokenizer function in mlpack?

19:14 < luffy1996> zoq I feel that policy gradient in complete from my side. It would be great if you go through the codes . I think you might like to merge this with the master branch.

19:16 < luffy1996> :)

19:18 < luffy1996> zoq: What are you plans to add bandit algorithms in mlpack. I believe it should be implemented for people willing to use mlpack for reinforcement learning :)

19:41 < travis-ci> mlpack/mlpack#4096 (master - 555557f : Ryan Curtin): The build has errored.

19:41 < travis-ci> Change view : https://github.com/mlpack/mlpack/compare/f465faf9d81a...555557f45a76

19:41 < travis-ci> Build details : https://travis-ci.org/mlpack/mlpack/builds/345664060

19:43 < manthan> hello, I am Manthan from India. I have a decent experience in Deep Learning and I am a DL/ML enthusiast. I wanted to ask if i can take up the task of implementing BatchNorm layer. Also if you could guide me on the use of pre-trained word vectors for string utilitites.

19:44 < manthan> @zoq @rcurtin : I have opened an issue already for the same.

19:55 < zoq> luffy1996: I'll take a look at the code once I get a chance. About Bandit algorithms, do you have something specifc in mind, before we add something we have to make sure there is a beneft, e.g. if we can come up with an implementation that is faster, or easier to use etc.

19:55 < zoq> manthan: Thanks for getting in touch, see my comment on the issue.

19:57 < zoq> kaushik_: There is no explit class for that, but depending on the task boost spirit could be helpful here.

19:57 < manthan> thanks, i am working on it.

19:58 < manthan> any guidance on the use of pre-trained word vectors/GLoVE for string to numeric representations?

21:03 < zoq> manthan: There is no existing GLoVE method yet that you could use, but there is an GSoC project idea that might change that: https://github.com/mlpack/mlpack/wiki/SummerOfCodeIdeas#string-processing-utilities