02:24 < zoq> Thanks for all the status updates, really interesting.

08:49 < haritha1313> zoq: You mentioned writing another class for merging two networks, similar to FAN. I am unable to find any such class. Could you please give some reference to it?

09:07 < ShikharJ> haritha1313: I guess it is Identity Layer, but again better to get confirmation from zoq!

09:17 < haritha1313> ShikharJ: Actually, I need to merge two networks (sequential/FNN) using add/multiply merge.

09:17 < haritha1313> By identity layer, you mean base layer, right?

09:20 < ShikharJ> I'm not sure if you can do this, but maybe you can try doing this (https://github.com/mlpack/mlpack/pull/1204/files#diff-73a3e2abae16e9a84a96856b191900e4R50)?

09:22 < haritha1313> Thanks for your help. I'll take a look at it :)

09:48 < jenkins-mlpack> Project docker mlpack nightly build build #333: SUCCESS in 2 hr 34 min: http://masterblaster.mlpack.org/job/docker%20mlpack%20nightly%20build/333/

14:09 < Atharva> Is it okay to initialise helper matrices while implementing functions for a class, inside the function?

15:25 < rcurtin> Atharva: I'm not sure I have enough context to say for sure, but I don't see any particular issue with that. I'd say, go ahead and do what you need to do, and if it's an issue it'll probably get brought up during a PR review :)

15:25 < Atharva> I figured it out, so no problem now.

15:40 < rcurtin> sounds good :)

16:06 < zoq> haritha1313: Actually I was talking about writing an extra class similair to the GAN class #1204, the idea is we can construct the GAN with two networks (generator and discriminator). We could do the same for th NCF module. Internally we merge the results. Do you think that could work here, if not let's just use Sequential layer, wich acts as a FFN class, but can be used with the merge/add layer.

16:11 < zoq> haritha1313: About the failing test, since it's just a failing simple test to see if we can retrain, let's use:

16:11 < zoq> arma::mat randomData = arma::zeros(100, 100);

16:11 < zoq> randomData.diag().ones();

16:11 < zoq> instead of

16:11 < zoq> arma::mat randomData(100, 100);

16:11 < zoq> randomData.randu();

16:11 < zoq> which should solve the issue.

16:23 < ShikharJ> zoq: Can I have a review of #1204 please? If I can be sure of its correctness for batchSize=1 case, then I can also run it for batchSize > 1 case and get the code merged.

16:48 < haritha1313> zoq: Wow, that just solved it :D . I still don't understand though why it was giving error for random values.

16:49 < haritha1313> Ah, you meant GAN. A typo then. Actually there wasn't much documentation available even for sequential layer. Thats why I thought of looking into the alternative solution you gave.

16:50 < haritha1313> I'll see how it turns out with sequential itself then.

19:47 < ShikharJ> zoq: Could you tell me how I could visualize or plot .arm data?

19:53 < zoq> one way ist to save the matrix as csv and to use pyplot or maplot

19:53 < rcurtin> if you were lazy, you could use the preprocessing utilities as a format converter:

19:54 < rcurtin> mlpack_preprocess_split -i data.arm -t train.csv -T test.csv; cat train.csv test.csv > data.csv

19:54 < rcurtin> (but that will shuffle the data too, if that matters)

19:54 < zoq> waht about using armadillo and save as pgm

20:02 < rcurtin> oh, yeah, that could work also

20:03 < rcurtin> but I think Armadillo will only do 8-bit PGM, which I guess will work if it's MNIST data or something

20:17 < Atharva> I am sorry if it's a very basic doubt, but what do delta matrices signify in the layer objects? When I studied backpropogation, I don't remember encountering it.

20:29 < Atharva> Isn't delta and error the same thing?