04:20 < travis-ci> mlpack/mlpack#4995 (master - 9d1f364 : Marcus Edel): The build passed.

04:20 < travis-ci> Change view : https://github.com/mlpack/mlpack/compare/949ab83ef5a3...9d1f364cc4c6

04:20 < travis-ci> Build details : https://travis-ci.org/mlpack/mlpack/builds/387602051

07:01 < ShikharJ> zoq: Supposedly I can't prepare the dataset on savannah.mlpack.org as numpy or pandas isn't available there. I'm preparing the dataset (albeit slowly) on my system, and I'll compress and scp into the server.

07:30 < Atharva> sumedhghaisas2: I opened the PR, I will add the backward function for KL in a few hours. Until then, if you are free, please take a look at the rest of the code.

07:31 < Atharva> sumedhghaisas: I think after this, I can start implementing the VAE class finally.

07:31 < Atharva> zoq: rcurtin: Sorry I didn't give any updates on the blog before this week, I will submit one today. :)

09:55 < jenkins-mlpack> Project docker mlpack nightly build build #339: SUCCESS in 2 hr 41 min: http://masterblaster.mlpack.org/job/docker%20mlpack%20nightly%20build/339/

10:05 < ShikharJ> zoq: I've prepared the dataset and pushed to the server, hopefully correctly this time. I'll debug the DCGAN test implementation for now.

13:51 < rcurtin> Github bought by Microsoft... unexpected news to start the week

13:56 < Atharva> Oh boy

15:25 < rcurtin> sumedhghaisas: sumedhghaisas2: I'm here now if you wanted to talk about that design issue, just let me know

16:58 < Atharva> rcurtin: It was about the GaussianDistribution class we have. It’s for multivariate distributions, right? It doesn’t have a standard deviation member. For the VAE class, we were thinking about using the distributions we have, but we needed to work with standard deviation. So, we wanted to ask your opinion about a class for independent univariate normal distributions.

17:01 < rcurtin> Atharva: right, the GaussianDistribution is for multivariate distributions, but if you need a one-dimensional distribution, then just pass in a mean vector of length one and a 1x1 covariance matrix

17:07 < Atharva> Oh, that was easy. Sorry about that. Just one thing, the 1 x 1 covariance matrix will be the variance of the one dimensional distribution, right?

17:11 < rcurtin> yes, it should be the variance, not the standard deviation

17:12 < Atharva> Okay, I googled it. It’s variance then

17:12 < Atharva> Oh haha

17:12 < rcurtin> yeah, sorry for the slow response... looked away for a bit :)

17:12 < Atharva> No it’s fine, I should have googled it anyway.

17:13 < Atharva> Also, I really think I need to revise statistics and learn some more maybe.

17:15 < rcurtin> it's always helpful :)

17:18 < Atharva> rcurtin: I was just curious, you have mentioned on your website that maintaining mlpack is one of the responsibilities that come with your job. So, do you give some of your work hours to mlpack?

17:18 < rcurtin> yeah---when I joined Symantec, the agreement was that part of my time would be reserved for research and mlpack, and that's continued to be true

17:18 < rcurtin> I'd like to see mlpack achieve some level of adoption inside the company, but speaking honestly, there are a lot of internal politics that have prevented success

17:19 < Atharva> Okay, the first part is really cool. About the politics, it’s a little sad.

17:20 < rcurtin> well, I don't think it's unusual for a big company

17:20 < rcurtin> it's possible that mlpack's deep learning toolkit may be used to do deep learning classification in the desktop antivirus product (Symantec Endpoint Protection)

17:20 < rcurtin> but the project is very slow moving... we started talking about it almost two years ago and still nothing has actually happened

17:21 < rcurtin> and even then, it will still be a difficult process to use mlpack for that, since basically every executive only knows about TensorFlow

17:21 < rcurtin> but hey, we'll see. I will keep writing code regardless :)

17:23 < Atharva> Oh, hopefully that project will happen. Also, it’s really cool to know that as a student I can contribute to something that can potentially reach so many people

17:23 < Atharva> I guess that’s the beauty of open source :)

17:26 < rcurtin> yeah, I hope that we can see more adoption in the future. I think the deep learning toolkit is really nice to use from C++

17:28 < Atharva> Yeah it is, I got used to it so quickly.