02:58 < KimSangYeon-DGU> ShikharJ: I saw your Mar 15th message, congratulations!! :)

04:04 < niteya> rcurtin: hey can please tell me what I should do with do my AdamW class wrapper , I really want to finish that PR.

06:33 < ShikharJ> KimSangYeon-DGU: Thanks!

07:33 < Viserion> What is the difference between Dense matrix and sparse matrix of user and items in collaborative filtering

07:56 < jenkins-mlpack2> Yippee, build fixed!

07:56 < jenkins-mlpack2> Project docker mlpack nightly build build #266: FIXED in 3 hr 42 min: http://ci.mlpack.org/job/docker%20mlpack%20nightly%20build/266/

10:29 < favre49> zoq: I mailed you a proposal, it would be great if you could give me some feedback when you get the time. Thanks :)

11:13 < sreenik> It's a bit of c++ code that I am having some trouble with. I have a function whose job is to choose (i.e. return) among a few layer classes, say Relu, Sigmoid, Selu and tanh based on the parameters I provide. So basically it has to return a class without knowing its type beforehand. Is there any way out?

11:15 < sreenik> I also cannot find any class that all of these inherit (thought about LayerTypes initially)

11:18 < zoq> sreenik: SFINAE might be helpful here: https://shaharmike.com/cpp/sfinae/

11:26 < sreenik_> zoq: thanks for the lead. I will just clarify a bit. Say I have a method declaration...

11:26 < sreenik_> template <typename T1, typename T2> FFN<T1, T2> foo(typename T::type) { cout << "1st" << endl; }

11:27 < sreenik_> I want to return FFN<SomeLoss<>, SomeInit<>>

11:27 < sreenik_> Is that possible?

11:30 < Viserion> Can anyone explain use and execution of statement in ReportIgnoredParam({{ "iteration_only_termination", true }}, "min_residue"); in cf

11:30 < Viserion> Can anyone explain use and execution of statement in ReportIgnoredParam({{ "iteration_only_termination", true }}, "min_residue"); in cf_main.hpp in collaborative filtering

11:34 < zoq> sreenik_: Yeah, the keyword you are searching for is decltype: https://www.cprogramming.com/c++11/c++11-auto-decltype-return-value-after-function.html

11:56 < Viserion> What is the significance of "min_residue" in ReportIgnoredParam({{ "iteration_only_termination", true }}, "min_residue");

12:12 < Viserion> What is this doing ReportIgnoredParam({{ "iteration_only_termination", true }}, "min_residue");?

12:59 < rcurtin> Viserion: have you tried reading the source and documentation for ReportIgnoredParam()?

14:00 < Nitish> @Ryan Curtin

14:01 < Nitish> sir , i want to write proposal for GSOC(project name =String Processing Utilities)

14:12 < shubhangi> I am trying to resolve 'https://github.com/mlpack/mlpack/issues/1772' this issue and I am getting this error 'https://paste.ee/p/wQc40' while running random_forest.py file in python debugger

14:36 < ShikharJ> shubhangi: Please read my comment from the previous day.

15:01 < rcurtin> Nitish: sounds good, go ahead

17:07 < riaash04> rcurtin: In the isomap pr you suggested to use minimum spanning tree to find all pair shortest path. In case of sparse graphs it is a good approximation but it might give some wrong results when graph is dense. Do you think it would be reasonable to use it? It would have lower time complexity than Dijkstra and Floyd's but since isomap depends on geodesic distance calculations sometimes it might produce wrong results.

17:08 < sreenik> zoq: thanks :)

17:14 < dinesh> can we load .mat files with mlpack

17:14 < rcurtin> riaash04: I thought that the MST would be exactly what's needed there, so I haven't managed to understand the discrepancy quite yet

17:14 < rcurtin> I'll take the time and leave more comments when I can, but it may be a little while

17:14 < rcurtin> dinesh: do you mean .mat files produced by Armadillo, or by matlab?

17:14 < dinesh> matlab

17:15 < rcurtin> ok, I see. I don't believe that Armadillo will directly import MATLAB matrices (you should double check their documentation), so I'd suggest writing your matlab matrix as CSV then importing like that

17:15 < rcurtin> or you could write the matlab matrix as HDF5 too, if you have HDF5 available on your system

17:16 < dinesh> in .mat files we can have more than 2 matrices in one file could we do that with CSv or HDF5

17:17 < rcurtin> huh, didn't know that

17:17 < rcurtin> for mlpack it'll only load one matrix out of a file

17:17 < rcurtin> same with Armadillo

17:17 < rcurtin> so you'd need to save multiple HDF5s or CSVs with one matrix in each I think

17:18 < dinesh> ok..thanks

17:19 < rcurtin> sure, happy to help

17:21 < riaash04> rcurtin: ok, thanks

17:25 < jeffinsam> zoq,recuritn : is there any way to print confusion matrix, i mean a function?

17:26 < rcurtin> jeffinsam: hmmm, I thought I remembered a function like this existing once

17:26 < rcurtin> but when I do 'grep -r Confusion src/' I don't see anything

17:26 < rcurtin> so perhaps no such function exists

17:27 < jeffinsam> recuritn: would it be apt to implement one to get recall and precision in one go ?

17:28 < rcurtin> jeffinsam: it could be useful; however, take a look at the cross-validation and hyper-parameter tuner code in src/mlpack/core/cv/ and src/mlpack/core/hpt/. These directories may have a good part of what you are looking for, and you may be able to reuse some of that code

17:30 < jeffinsam> rcurtin : thanks

17:33 < rcurtin> jeffinsam: sure, happy to help

17:34 < rcurtin> I think I saw you opened some PRs... I'll try and review when I have a chance, but there are a lot of PRs and I'm pretty underwater with reviews right now

17:34 < rcurtin> so it might be a little while :)

17:34 < rcurtin> (at least from my end. someone else may get to it before I do, and I think actually Shikhar might have already, if I'm remembering right from my quick glance at emails this morning)

17:35 < jeffinsam> rcurtin : no issues, should i write some test for it, update cmake, to check wether there is issues with the code

17:36 < rcurtin> yeah, definitely any new functionality should have some tests to make sure it works right :)

17:37 < jeffinsam> recutin : also was working on dictionary encoding idea , i have some doubts, Like if there are multiple document : for exampel - [ hi hello how are you ] [ i am good] .... then ecoding vector may have 2 different sizes

17:37 < jeffinsam> so should i pad 0 or just leave it to user to handle it

17:38 < jeffinsam> padding 0 would add zero-risk bias

17:44 < rcurtin> if the API is such that a user passes in a single document, then no need for padding; but the user can pass in multiple documents then the encoding vector must be able to fit all words in the dictionary across all documents

17:47 < jeffinsam> rcurtin : umm then i must padd it with 0 i guess.

17:48 < jeffinsam> ok so i would try to code that too.

17:50 < sreenik> Is there any superclass that all the loss type classes inherit, so that in a function I may have return type as the superclass type and then return a particular loss (say MeanSquaredError)?

18:00 < zoq> sreenik: You could use LayerTypes.

18:01 < sreenik> Sounds good. I will give it a try

20:00 < Viserion> I would like to work on issue for adding support for alternative normalization...how should I begin?

20:03 < zoq> Viserion: Each normalization for cf follows a common interface, so you should implement that.

20:08 < Viserion> First thing I thought about is to get the normalization algo, which we will get from input string alongside decomposition algorithm..

20:34 < Viserion> Why we are making copy of data before normalizing it, in cf_impl?

20:58 < Viserion> @zoq?

21:12 < zoq> Viserion: To keep the input data untouched.

21:15 < zoq> Viserion: Normalize takes the input data as a reference.

21:17 < Viserion> I basicaly changed the functions to get the normalization algo in the related files...how can I verify them?

21:18 < zoq> You can run the cf test suite: 'mlpack_test -t CFTest'

21:18 < zoq> For more details see: https://github.com/mlpack/mlpack/wiki/Testing-Guidelines