01:49 < zoq> ShikharJ: I think Kris used the example code to produce the weights, but I'm not sure.

02:16 < zoq> ShikharJ: in gan_impl.hpp generator.Gradient(generator.parameter, numFunctions, gradientGenerator, batchSize); calls the FFN Gradient function with numFunctions which is 500 in case of the mnist test, but this should be noise.n_cols, also instead of setting currentInput = noise, we should set predictors since this is what is going to be used inside the FFN Gradient/Evaluate function.

02:16 < zoq> ShikharJ: I hope anything I said makes sense, let me know if I should clarify anything.

02:35 < rcurtin> zoq: sorry for your inbox, I thought I had that job configured to only send one email on an unstable build!

02:38 < rcurtin> I think it will still send ~50 more as the builds become stable, but I edited the job in the future to use a different email notification plugin that should only send one email

02:38 < zoq> rcurtin: Are we talking about the 'docker mlpack nightly build'?

02:46 < zoq> I got two mails yesterday, which is absolutely fine.

03:21 < rcurtin> the weekly build emails seem to be going out now

03:21 < rcurtin> I think I have something like 100 so far...

06:35 < haritha1313> zoq: What is the right way to use the merge layers to connect two networks? I am able to merge two single layers using add/multiply merge. But I have got two (Lookup+Linear) networks, whose output needs to be merged.

06:35 < haritha1313> I am not able to find much in the documentation either of add merge.

06:48 < ShikharJ> zoq: preTrainGAN is basically a parameter to specify how many images should be pre-trained on the discriminator before beginning the adversarial routines (using both generator and discriminator side by side).

06:53 < ShikharJ> numFunctions is basically the total number of images available in the dataset.

07:02 < ShikharJ> zoq: Ah, ignore my message above, I got confused between discriminatorPreTrain and preTrainedGAN. I wouldn't have an idea about the latter, but the test in whoch that is used is not a part of O'Reilly test. Maybe we can ask lozhnikov?

07:02 < ShikharJ> *which

07:16 < ShikharJ> zoq: I agree that we should use noise.n_cols in place of numFunctions. I'll make the changes.

07:18 < ShikharJ> zoq: I don't seem to grasp why we should set currentInput for generator to predictors (we are doing that for the discriminator already). Could you explain that?

07:19 < ShikharJ> (Assuming all of this has to be done inside GAN::Evaluate function).

09:56 < jenkins-mlpack> Project docker mlpack nightly build build #329: SUCCESS in 2 hr 42 min: http://masterblaster.mlpack.org/job/docker%20mlpack%20nightly%20build/329/

11:05 < lozhnikov> ShikharJ: If I remember right, GANTest is an attempt to sample gaussian 1d distribution (see section "Approximating a 1D Gaussian distribution" http://blog.aylien.com/introduction-generative-adversarial-networks-code-tensorflow/)

11:08 < ShikharJ> lozhnikov: Ah, yes I remember you mentioning it before, thanks for reminding me :)

11:09 < lozhnikov> not at all

14:20 < manish7294> rcurtin: Can you please re-send me the server credentials, specifically hostname? Accidently, I lost them. :(

14:29 < sumedhghaisas> Atharva: Hey Atharva

14:29 < sumedhghaisas> how is it going?

14:30 < Atharva> sumedhghaisas: Hi Sumedh

14:35 < sumedhghaisas> Atharva: How were your holidays? Got your mail regarding KL divergence. Excited to get the project started.

14:39 < rcurtin> manish7294: slake.mlpack.org :)

14:39 < rcurtin> Atharva: hello, hope your vacation went great! :)

14:40 < Atharva> sumedhghaisas: rcurtin: It was amazing. Europe is beautiful

14:41 < Atharva> Yeah, I am excited about the project as well.

14:45 < Atharva> I am working on the KL divergence part right now

14:49 < sumedhghaisas> Atharva: That sounds good. Just wanted to have a quick chat about the communication aspect of things so we get that out of the way. Do you prefer setting up a specific time of the week to have a weekly sync?

14:52 < Atharva> sumedhghaisas: Yeah I think we need to do that, I am at my home the whole time, so it’s upto you. What time of the week are you free?

14:54 < sumedhghaisas> Atharva: I prefer Thursday. But let me know if that doesn't work for you and we can arrange for something.

14:54 < sumedhghaisas> So we have a time difference of 4 30 hours now. huh...

14:54 < sumedhghaisas> what time of the day do you feel comfortable?

14:55 < Atharva> sumedhghaisas: Thrusday is fine. Yeah we have that time difference

14:56 < Atharva> I sleep late at night, so maybe 9 british time?

14:57 < Atharva> Or anytime really, as I said, I will be at home all the time

14:59 < Atharva> sumedhghaisas: What’s good for you?

17:45 < ShikharJ> zoq: I have made the required changes, but it still seems that the error in generator.Gradient persists. I'll fix it up over the weekend, and get started with the DCGAN API.

18:26 < travis-ci> ShikharJ/mlpack#159 (GAN - 291564c : Shikhar Jaiswal): The build has errored.

18:26 < travis-ci> Change view : https://github.com/ShikharJ/mlpack/compare/369ace5c8e19...291564cf8a70

18:26 < travis-ci> Build details : https://travis-ci.org/ShikharJ/mlpack/builds/383812881