01:09 < travis-ci> mlpack/mlpack#1774 (master - 276e268 : Marcus Edel): The build passed.

01:09 < travis-ci> Change view : https://github.com/mlpack/mlpack/compare/064653df2890...276e268db9f8

01:09 < travis-ci> Build details : https://travis-ci.org/mlpack/mlpack/builds/197851638

16:57 < dineshraj01> Hi guys, I am really having a hard time understanding mlpack code. Generic programming is used extensively, but I don't seem to understand what each and every typename is and what it does. Although I am reading the comments in the code, but still not sure about the complete structure. Any tips as to what approach I should use to better understand the code?

16:59 < rcurtin> hi dineshraj01, there are typically three techniques that are used:

16:59 < rcurtin> * SFINAE (substitution failure is not an error)

17:00 < rcurtin> * policy-based design (splitting each class into a bunch of template parameters)

17:00 < rcurtin> * class traits

17:00 < rcurtin> you are lucky, I just found some tutorial links for each of those things:

17:00 < rcurtin> http://shaharmike.com/cpp/sfinae/

17:00 < rcurtin> http://www.drdobbs.com/policy-based-design-in-the-real-world/184401861

17:00 < rcurtin> https://accu.org/index.php/journals/442

17:01 < rcurtin> on the other hand, I can also give you a quick answer for the code you are looking at :)

17:09 < dineshraj01> thanks @rcurtin for sharing the links

17:10 < rcurtin> sure, I hope they are helpful

17:18 < dineshraj01> yeah and actually, I was lookiing at the ns_model in neighbour search wherein boost::variant has been used with visitor and an issue on github says, the same needs to be done for the rsmodel and ramodel. Since, there were so many parameters like TreeType, TreeStatType,tree traversals and others. Since I couldn't visulaize the overall flow, I was just blank minded as to how to refactor the code.

17:22 < dineshraj01> and what exactly are these models in range_search, neighbour search??? comments say it provides easy way to searlize the model itself, abstracts away the model...but i still don't get it :(

17:27 < rcurtin> I'm not sure what you mean by "these models"

17:27 < rcurtin> do you mean NSModel?

17:27 < rcurtin> or RSModel

17:27 < rcurtin> the reason those classes exist is that it's not clear how to provide a unified interface for serializing models of different types

17:27 < rcurtin> the command-line programs can have use many different NeighborSearch or RangeSearch types, because of the many template parameters

17:28 < rcurtin> so the NSModel class exists to encapsulate all of these types inside of a single class that can be serialized

17:28 < rcurtin> and then from the command-line program we can always serialize an NSModel instance, and we don't have to decide which NeighborSearch<...> type to serialize

17:59 < dinesh_> okay. thanks for the explanation @rcurtin

20:13 < rcurtin> I found a nice writeup by an mlpack user today on getting mlpack set up on Ubuntu: http://mccormickml.com/2017/02/01/getting-started-with-mlpack/

21:25 < zoq> Really nice tutorial, there is one part I wouldn't agree on "it appears that mlpack has a lot of dependencies on, e.g., other linear algebra libraries"