02:42 < heisenbug_> hey zoq, rcurtin...can I get some feedback on my proposal? How to improve it for next time?

02:43 < heisenbug_> https://docs.google.com/document/d/1cMc1BLdzUS1x2qICx34m3B9Qhc4nfj7H6XN31W9baaY/edit?usp=sharing

07:19 < mulx10> jeffin143: Yes, I tried with a slight variation. (https://pastebin.com/wmsiGMrz)

07:51 < jeffin143> Mulx10 : I ll look at it once I reach home, travelling now

07:51 < jeffin143> Also if it works , I think it's good idea to make a tutorial and post it either on docs or mlpack /models

07:51 < jeffin143> Bcse I guess not many would now that such thing could be done and hence reference could always be good, may be I could be wrong

08:04 < jenkins-mlpack2> Project docker mlpack nightly build build #320: STILL UNSTABLE in 3 hr 50 min: http://ci.mlpack.org/job/docker%20mlpack%20nightly%20build/320/

08:45 < mulx10> https://pastebin.com/swh1kzFG this could be used to load a folder. the final mat contains data in the form that could be directly fed into the network.

12:15 < Mulx10> jeffin143 : yes sure

12:28 < rcurtin> Mulx10: looks nice. for the second loop, you could probably get some speedup for this in two ways:

12:28 < rcurtin> (1) allocate the full size of dataset instead of calling join_cols() repeatedly (each call to join_cols() will copy the matrix)

12:29 < rcurtin> (2) use 'false' as a fourth parameter to the constructor of `arma_mat` to avoid copying the memory, so arma_mat is just an alias around the memory held by opencv_mat

12:29 < rcurtin> in any case, that whole snippet is nice and short and easy to understand, which is great :)

13:28 < Mulx10> rcurtin : thanks, should I start implementing it?

13:36 < rcurtin> Mulx10: up to you, I was just throwing some suggestions out there :)

15:14 < Toshal> ShikharJ: saksham189: Hi may be somewhat late. I am highly sorry for that. I will be there till 9:15

15:35 < ShikharJ> Toshal: Alright, no problem, please let us know when you're here.

15:51 < ShikharJ> saksham189: I think we should start now. Toshal can look up the IRC logs for the previous conversation.

15:51 < saksham189> alright

15:52 < ShikharJ> Okay, so first of all congrats on getting selected :) We had 80+ proposals this time, and merely 9 slots.

15:53 < ShikharJ> I'll be mentoring you and Toshal for the summers, but feel free to ask other mentors for help, and make conversation with everyone :)

15:54 < saksham189> thanks :) It's great to be part of mlpack.

15:55 < ShikharJ> We'll have a another formal introduction sometime next week (Marcus is overseeing that). There, we'll introduce you both to the broader mlpack community and the story behind mlpack since its inception and the people who have worked on it (and a few success stories of mlpack as well).

15:56 < ShikharJ> For now I think, we should start up by deciding on a time for bi or tri-weekly updates. Which one is more preferable to you?

15:57 < saksham189> I think that bi-weekly is good. Although we can give any urgent updates through IRC anytime.

15:58 < ShikharJ> Okay great, Toshal please provide your preference as well. Also, if you have any other plans for the summer (vacation / work) please let me know. We'll need to decide for the days and timings as well, since out timezones would be different.

16:01 < saksham189> I don't have any other plans for summer as of now. I will be moving to Vietnam for summer as my parents live there.

16:01 < ShikharJ> saksham189: What about Monday and Thursday / Friday? We can have meetup in the IRC channel itself.

16:01 < Toshal> Hi, I am also fine with bi-weekly update.

16:02 < saksham189> I am fine with any day.

16:02 < ShikharJ> saksham189: Whoa, interesting country. Have fun! Let me know when you need the time off for traveling and vacations :)

16:02 < ShikharJ> Toshal: And what about the days?

16:02 < Toshal> ShikharJ: can't we meet on weekends?

16:03 < Toshal> I am highly free on those days.

16:03 < ShikharJ> Toshal: Do you have any plans for the summers? I don't think I would be available much on weekends.

16:04 < Toshal> I also don't have any other plan this summer.

16:04 < Toshal> Except some schools and papers.

16:04 < ShikharJ> Oh yeah, please let me know of your college re-opening schedules.

16:05 < Toshal> Okay Friday is fine for me.

16:06 < saksham189> ShikharJ: my college would open around 1st August.

16:06 < ShikharJ> Toshal: I was hoping the meeting would be more spread out over the week. There'd be literally nothing to talk about on Sunday, after we've met on Saturdays right?

16:07 < Toshal> ShikharJ: My college re-opening may be on 1st June or 15th June. But, it will be only for one month.

16:08 < Toshal> ShikharJ: Yes, as per your expertise. I am fine with any time.

16:08 < ShikharJ> saksham189: Okay, great so I think we have a good date range there. Let me know if you want me to shift your coding time dates to suit your needs, since you'll be busy with college in August. You can cover up for that time in this month if you feel okay with that.

16:09 < ShikharJ> saksham189: That's assuming you're free to work now, and willing to start soon, but no issues there.

16:09 < Toshal> ShikharJ: I don't need to attend college that much but will need to be in touch with it.

16:09 < ShikharJ> Toshal: Hmm, 1st June? Damn, you're working too hard :)

16:10 < Toshal> Yeah, university and some college rules.

16:11 < ShikharJ> saksham189: Can we decide on a time? For this month, I'll be primarily in India, so I think post 9pm should be suitable for everybody? Let me know if any of you have any objections for that?

16:12 < ShikharJ> Monday and Friday post 9pm?

16:12 < saksham189> yes thats good for me.

16:13 < Toshal> Fine with me as well

16:13 < ShikharJ> Coo, we can decide for the next month later. Next up, I think we need to decide on the work that needs to be done between the three of us.

16:14 < ShikharJ> I had the broad desire to include the GAN framework in the next release, which will probably happen towards the end of August.

16:14 < Toshal> Yes me too.

16:15 < ShikharJ> I see that you both are already acquainted with the coding conventions and are already familiar with the ANN code, so that works in our favour.

16:16 < ShikharJ> For now, let's aim for finishing up on the pending work till the Community bonding period ends? If anyone wants to take a break for now and start in the coding period itself, now is the time to speak up :)

16:17 < ShikharJ> By pending work, I mean the pending PRs, that are currently open.

16:18 < saksham189> yes I think that it would be good to have them finished and merged.

16:19 < ShikharJ> For saksham189 I see two related to GANs and one related to Highway networks. Both of the GAN related ones are nearly complete, not sure of the highway networks PR, please enlighten us on that.

16:19 < Toshal> I am fine. Just as I mentioned I may not be in that great touch between 17 to 28 may due to exams. But I will be in touch for sure.

16:20 < ShikharJ> For Toshal I see GAN serialization and Dual Optimizer. For now I think we should focus on Serialization, Dual Optimizer is going to be a big PR in my opinion, so we can leave that for the coding period.

16:21 < saksham189> ShikharJ: I think it is close to completion. I have added the tests and documentation. Further reviews would be helpful.

16:22 < ShikharJ> saksham189: Okay great, also, I think that one of the PR had merge conflicts, so that might need to get sorted out.

16:22 < Toshal> ShikharJ: Okay, I guess I can finish dual optimizer implementation before my exam begins. Testing looks like tough nut. Also I will mention quite important point after saksham finishes.

16:22 < saksham189> ShikharJ: yes thanks for the reminder. I'll do that tonight

16:23 < ShikharJ> My work for now would be to provide reviews on your current PRs and try finishing up my own implementation of CycleGAN. I've left that one neglected for a long time now :)

16:24 < ShikharJ> Okay, after this all is done, we'll decide on splitting up the work where your proposals overlap. One such point would be StackedGAN I think, but that is all in the future. So for now we're good.

16:26 < Toshal> ShikharJ: I have some important points. I am starting them right now.

16:26 < ShikharJ> I hope that this experience is enjoyable for both of you, full of learning and fun. We're going to have a great summer, and all the mentors will be there for help.

16:26 < ShikharJ> Toshal: Sure, go ahead.

16:27 < Toshal> 1. Orielly test is quite a lot big. It will take quite a lot of time to test it. With my estimation approx 50 days.

16:27 < Toshal> It's currently running.

16:28 < ShikharJ> Toshal: Where are you running it? On what machine?

16:28 < Toshal> Today is the 7th day. I am running on i5 8th gen.

16:28 < saksham189> Toshal: are you trying to run using the full mnist dataset?

16:29 < Toshal> ShikharJ: yes I am running on full mnist dataset.

16:29 < ShikharJ> Toshal: Are you running on your own machine?

16:29 < Toshal> ShikharJ: Yes, I am running on my own machine

16:30 < ShikharJ> Toshal: I'd advise you not to do that. We have an online heavy-duty node for running these tests. We'll be providing you access to that in the coming days. That should help greatly.

16:31 < Toshal> Okay.

16:31 < ShikharJ> Toshal: It's where I ran my tests, full MNIST under 7 hours on single core.

16:31 < ShikharJ> Toshal: Oreilly took about 4.5 hours on multiple cores (single core aggregate was around 11 hours).

16:32 < ShikharJ> So testing should become a lot less painful in the days to come. Till then, I'll run your code on the server, and cross-check the output. Sounds good?

16:34 < Toshal> 2. Dual optimizer are quite tough for WGAN-GP. We may need to modify current FFN implementation quite a lot. In WGAN-GP we need to train two batches on one loss and the pseudo batch on another loss. Currently, FFN implementation only support one loss so that might be somewhat painful.

16:35 < ShikharJ> Toshal: I am yet to review your comment on the PR regarding the Dual Optimizer for WGAN-GP. I'll get to that when I have some time, and then maybe we can proceed from there?

16:41 < ShikharJ> saksham189: Okay, so I think, we're done for now. See you later. Toshal has disconnected, so probably, he's also finished. Have a good night!

16:41 < saksham189> ShikharJ: yes, thanks for your time!

16:41 < ShikharJ> We'll have a quick chat on Monday then.

16:43 < Toshal> ShikharJ: Sorry I just got disconnected.

16:43 < Toshal> Okay. Regarding the serialization thing. I would say, a lot of code would be changed after Dual Optimizer's gets implemented. So I was thinking that │

16:43 < Toshal> │ | lets finish with Dual Optimizer.Let

16:44 < ShikharJ> Toshal: Okay, please suit your preferences, we'll take up Dual Optimizer for now then.

16:46 < Toshal> ShikharJ: Thanks, that got creepy. :)

16:48 < Toshal> saksham189: Just in the end. Can I get a copy of your proposal. I would share mine too. It will help us to know where we will be coinciding.

16:50 < saksham189> Toshal: yes we can do that.

16:51 < Toshal> Okay I will be sending it to your personal mail soon.

16:54 < Toshal> ShikharJ: saksham189: Thanks for your time.

16:55 < Toshal> And sorry for being late.

16:55 < ShikharJ> No worries

17:18 < Toshal> rcurtin: zoq: Sorry for disturbance. This is just a reminder. It want to remind of https://github.com/mlpack/ensmallen/pull/109.

17:39 < rcurtin> Toshal: no problem. I am working through my backlog of PRs at the rate of 3 a day :) I'll try and get this one soon