02:51 < rajeshdm9> Hey guys, Rajesh here. @zoq, you had asked me to try out implementing my own environment for mlpack. I am currently working on that. For testing the correctness should I run a agent with some policy in this environment ? I am still figuring out how more about mlpack so not sure about this - so is that the way or am I missing something ?

06:05 < sshkhrnwbie> @rcurtin : Could you please go through PR #1241 and suggest me useful tests when you have some time. Once this gets closed I can get started on he and glorot initializers. Thanks

06:19 < raden> Hi all.

06:19 < raden> I am interested to participate GSoC 2018 as student. How do I start

06:41 < Atharva> raden: Welcome to mlpack! :) To get started try building mlpack from source and get comfortable the codebase. Then you can look into this issue #1152.

12:19 < rajeshdm9> Hey guys, Rajesh here. @zoq, you had asked me to try out implementing my own environment for mlpack. I am currently working on that. For testing the correctness should I run a agent with some policy in this environment ? I am still figuring out how more about mlpack so not sure about this - so is that the way or am I missing something ?

12:59 < d4rth_> hey sumedhghaisas, i am interested in the variational autoencoder idea, any headsup?

12:59 < d4rth_> looks fairly high prio to me, is it fair to pursue it this year?

13:00 < d4rth_> sorr yfor being abrupt, i am karan, undergrad at iit roorkee and a past gsoc student with opencv

13:06 < sumedhghaisas> d4rth_: Hey Karan. Welcome to MLpack.

13:07 < sumedhghaisas> VAE project is up for this summer.

13:08 < d4rth> Cool, I will quickly have a look at the tickets and get back.

13:08 < d4rth> Thanks!

13:08 < sumedhghaisas> if you have any questions regarding the objectives or requirements, let me know

13:08 < d4rth> Sure.

13:08 < sumedhghaisas> ahh there are no open tickets for this project yet.

13:09 < sumedhghaisas> you will be starting a framework in ANN

13:09 < d4rth> Oh yes I tried a keyword seach :(

13:09 < d4rth> Anyhow it sounds interesting

13:10 < sumedhghaisas> have you worked on generative models before?

13:10 < d4rth> I have a bit of experience with seq2seq models.

13:10 < sumedhghaisas> I have attached the relevant papers in the project description.

13:11 < d4rth> I have read Kingma's paper. I will read through the rest of them.

13:12 < sumedhghaisas> Sure. Give them a try. VAE model is not very complex in my opinion. although some thought must be given to the loss function.

13:14 < d4rth> Great. I will have a look at the papers then, I'm here meanwhile.

13:16 < sumedhghaisas> Great!

14:18 < satyam_2401> @zoq: i read the pr on pso. It is different than what i wish to implement. So, i have cloned the master repo, and I will try my implementation and see how it turns out. So, either you can monitor the cloned repo or else i will send a pr request. Let me know what you think. And we can keep discussing improvements in algorithm here. I wish to discuss "association and niche preservation method" for selecting global

14:18 < satyam_2401> guide, i think it has not been implemented yet in the literature.

15:15 < ShikharJ> Looks like Appveyor has been stuck for a while now.

15:17 < moksh> Hey @zoq, I tried running the gym tcp api locally, I am facing some errors as I have gym installed on python3, anyway I can configure the API to work with python3?

15:28 < rcurtin> ShikharJ: I cancelled the oldest queued job and the next one looks like it started successfully

15:29 < rcurtin> I guess it will take a little while to work through the backlog

15:32 < ShikharJ> Can confirm, it's running well now. I thought it might've been a server-wide issue.

18:14 < Sayan98> Hello everyone, this is Sayan. I would love to contribute to mlpack and work on the project on Reinforcement Learning (https://github.com/mlpack/mlpack/wiki/SummerOfCodeIdeas#reinforcement-learning). I am building mlpack from source as we speak. This is my first message on this channel so please point out if I make mistakes. Looking forward to be a part of the mlpack family. Thank you!

18:16 < zoq_> Sayan98: Hello and welcome, there are some discussions on the mailing list about the RL project, so it's probably a good idea to check them out.

18:18 < zoq_> satyam_2401: If you agree, let's open a PR, makes it easier to discuss details.

18:21 < zoq_> moksh: Oh, would be awsome if you could open a PR with the necessary fixes.

18:24 < zoq_> rajeshdm9: Yeah, testing an environment is tricky, since even if we can solve the task using e.g. QLearning, it's not granted the environment is correct. But I think this is still bettern than nothing. Anyway, I guess depending on the environment there might be additional options we could do like testing edge cases.

18:24 < moksh> zoq: I am not really familiar with Elixir, but I'll try and get back to you, thanks.

18:29 < zoq_> moksh: Okay, let me know if I should clarify anything. I can also try to reproduce the issue.

20:28 < satyam_2401> @zoq: sure. Are we targeting only single objective pso?

20:30 < satyam_2401> since the pr you mentioned about has implemented single objective pso, and the statement here, " First implement one or two unconstrained methods and afterwards takes a look at one or two constrained methods." doesn't explicitly says whether it is single obj. or multiple obj.

20:31 < zoq> satyam_2401: At least I would start with it, but it would be great if we could implement a multiobjective optimization method.

20:31 < satyam_2401> I was thinking about multi obj. pso until now.

20:32 < zoq> satyam_2401: I guess in this case, why not just go for it :)

20:32 < satyam_2401> The problem in selection of global guide lies in mopso since it might not cover all the solutions.

20:33 < satyam_2401> okay, I will implement a multi objective pso

20:33 < satyam_2401> and then send a pr

20:34 < zoq> Awesome :)

20:34 < satyam_2401> I have written multi objective pso before, but it has different coding style

20:34 < satyam_2401> I have not used template metaprogramming

20:34 < satyam_2401> So, i will try to convert into mlpack's interface and send a pr then.

20:34 < rcurtin> the IPs for the build systems are supposedly being assigned today in 90 minutes

20:35 < rcurtin> so... maybe they will be up tonight. I'm pretty frustrated, we are getting close to two weeks of downtime

20:36 < zoq> satyam_2401: If you like you can start with the implementation and open the PR, I can help you with the transition.

20:37 < satyam_2401> That would be great. Thanks.

20:37 < zoq> rcurtin: Fingers crossed, waiting for the Static Code Analysis job :)

20:37 < rcurtin> yeah really... there are lots of PRs waiting...

20:38 < zoq> satyam_2401: This is a neat project to get familiar with the codebase.

20:38 < rcurtin> if they do not get things online today, then I have a call with the team tomorrow around lunchtime and I will yell at them

20:38 < rcurtin> I probably won't actually yell... I guess I am more likely to desperately plead with them...

20:38 < zoq> :)

20:44 < zoq> I think the problem is, there isn't much we can do about it ... I don't think for them it's crucial to get everything back online as soon as possible, and if it's on the top of the list they have some problems.

20:45 < rcurtin> exactly

20:52 < zoq> https://mobile.twitter.com/smdiehl/status/964136884019630082 maybe we should do a workshop :)

20:54 < rcurtin> heh, I think that is the truth :)

20:54 < rcurtin> shogun did a workshop with numfocus; I think it might be cool to colocate one at some point

20:54 < rcurtin> you went to that one I think?

20:55 < zoq> yeah, it was more like the left image :)

20:56 < rcurtin> in order to actually run one of those, I think maybe we would need to join NumFocus

20:56 < rcurtin> if you'd like, I can send Heiko an email and start a more serious discussion

20:57 < rcurtin> I don't see any particular reason to avoid joining numfocus; I don't think they'll assert any control over the project or anything

20:57 < rcurtin> there are a lot of big-name projects that are sponsored by numfocus now (including numpy, pandas, matplotlib, and julia)

20:57 < zoq> I still want to see what other options are.

20:58 < rcurtin> ok, that sounds good to me

20:59 < rcurtin> if you want to take a look and see what else is out there and let me know what you find, we can figure out what to do

20:59 < rcurtin> and inaction is an option too---I don't see a desperate need to become a part of a sponsoring organization (other than possibly getting funding to put together a workshop or something)

20:59 < zoq> right