11:06 < tarun_ml_enthusi> hello naywhayare...im sorry i had exams for the past two weeks and wasnt able to find any time...

11:06 < tarun_ml_enthusi> im free now

11:08 < tarun_ml_enthusi> and eh if you remember i was strugling to get started off...still am actually....i looked into bug #372 ..read all about CF and tried to find as much info as i could about matrix factorization and then its use in CF

11:08 < tarun_ml_enthusi> could you tell me a lil more in detail about the bug..or what i need to do to solve it?

11:14 < tarun_ml_enthusi> im sorry for testung your patience again and again...but i really want to contribute and thats why after not understanding shit or while facing trouble i keep coming back here

13:13 < qwebirc918773> @naywhayare ....in the bug #372 , ive understood most of it but require some clarity in some parts

13:14 < tarun_ml_enthusi> yeah so basically NMF ALS is used to minimize the regularized SVD right?

13:15 < tarun_ml_enthusi> and we use regSVD cause the data might be incomplete right?

13:16 < tarun_ml_enthusi> so if NMF ALS is taking too long , we could try using SGD i.e stochastic gradient descent

13:17 < tarun_ml_enthusi> i know that ALS has some advantages over it when it comes to implicit data , but SGD is supposed to be faster

13:17 < tarun_ml_enthusi> and maybe the algorithm is not making use of the fully parallelization ability of ALS

13:18 < tarun_ml_enthusi> i dont have a clear understanding on ALS...but do you think its worth a shot trying out SGD?

13:21 < tarun_ml_enthusi> i had to do alot of research and reading right from scratch to understand what i have until now...just the theory and whats going on ...to actually make changes to the code and try out a new algo , ill need alot of help and guidance from you :)

15:36 < naywhayare> tarun_ml_enthusi: I hope your exams went well :)

15:37 < naywhayare> for #372 the things to do are to first play around with the tolerance of the NMF-ALS algorithm; maybe it is way too tight, causing the test to take so long

15:37 < naywhayare> the next thing to do is to use gprof to profile the test, to see what parts of it are taking the longest

15:37 < naywhayare> if the tolerance is too tight, or if the profiling shows that there is a bottleneck that can be fixed, we can do that

15:38 < naywhayare> otherwise, we can consider changing the default factorizer, like Siddharth suggested