02:44 < mjscott> rcurtin: in response to the sparse SVD stuff, I can't guarantee that all rows/columns have entries

02:46 < mjscott> however, the density of W and H is about the same as V. Like, it is simply not training properly I think, and I'm not sure why. I'm using the SparseSVDCompleteIncrementalLearner, since gradient descent will be faster than alternating. Even though the residue goes to 1e-5, All of the predictions are about the same.

02:49 < mjscott> like, all of the entries in V are in the range 1-5, however, when I factor V with rank 100, I do dot(W.row(i), H.col(j)) for prediction of V(i, j), and for every prediction I get around 20-23, which is way over the range of the entries in V.

11:39 < Trion> Thanks vss and zoq for the appreciation :-)

11:59 < Trion> How will gym_tcp's cpp files be added as dependency in agents?

12:21 < zoq> Trion: I thought since it's header only, we can write a simple cmake subroutine that fetches the code and that's it.

12:30 < Trion> Nice, that would make the folders much cleaner. I think I should learn cmake :P

15:56 < Trion> In the Copyright.txt I saw CodeAI, did it add its name in there by itself too? :O

16:00 < zoq> :) I guess you could learn that behaviour from the commits, but in this case no: https://github.com/mlpack/mlpack/commit/03ba9bf6d1bb047bb83e1359842b42ef7481d66b#diff-333ea6639142920a50d487bef1bfb724

16:02 < rcurtin> they told me they would send more PRs but they haven't done it yet :)

16:10 < Trion> :P Nice! AI is not going to take Coding jobs just yet