12:09 < nilay> zoq tham: i have completed feature extraction part. i am pasting the gist here. i will convert this in mlpack format after some time.

12:15 < nilay> i am getting some error.

12:15 < nilay> i'll post in some time, sorry.

14:21 < mentekid> rcurtin: I have some "alternative" code to the one you included in your PR - how do I share it with you? Can I commit to that PR in some way, should I open a different one?

14:26 < rcurtin> mentekid: the way I did it for your PR was to check out your PR branch and then commit my code

14:26 < rcurtin> then use git format-patch to create the patch and send it to you

14:27 < rcurtin> maybe there is an easier way bit that way at least works... :)

14:59 < mentekid> I emailed it to you :)

15:00 < mentekid> the attachment doesn't show up at the github interface though :/

16:08 < rcurtin> mentekid: thanks, I will merge it in when I can

16:08 < rcurtin> probably not today, I am at the track racing karts :)

16:17 < mentekid> sounds fun! It's not really ready for a merge yet, if we end up going that way we need to change the serialization and probably other interface things

16:38 < rcurtin> wow, a comment from Sterling, h

16:38 < rcurtin> I had no idea he still watched the github repo :)

18:02 < keonkim> cli executables seems to have two major styles of writing. The one that initializes every parameters at the beginning of main() and one that initiailizes parameters right before they are needed. Which one should I go for as I write new executable?

18:03 < keonkim> oh, and sometimes there is also two kinds of checks, which really does the same thing: CLI::HasParam() and parameter.empty(). which one is better?

18:04 < keonkim> (I mean preferred)

18:10 < keonkim> one that is most comfortable to my eyes is initializing everything at the beginning and checking by CLI::HasParams.

18:20 < keonkim> I want to unify the styles while going through every _main.cpp s

18:24 < TD> To install mlpack on Windows 10 do you need to still install Armadillo, ProBoost, Etc;

18:24 < TD> Currently, I have downloaded the mlpack from the appveyor site

18:25 < TD> I have copied and pasted the files into my VS2013 solution explorer

18:26 < keonkim> I just went through that like yesterday. I might be able to help you.

18:27 < TD> if you could it would be greatly appreciated

18:28 < keonkim> my version was VS2015. I installed using nuget.

18:28 < keonkim> you can just follow the instructions in https://github.com/mlpack/mlpack/blob/master/.appveyor.yml

18:29 < TD> add that is it?

18:30 < keonkim> ya Let me get back to my windows machine, and post detailed instructions on my blog really quick.

18:31 < keonkim> will take about 30 min.

18:31 < TD> K, cool and I greatly appreciate this.

18:32 < keonkim> no problem, it was in my todo list anyways. :)

18:32 < TD> I have spent the last 7 days trying to get this to work and I think I have increased mlpack.org site visibility by 100% :-)

18:42 < keonkim> as I think of it now it might take more than 30 min, 1 hr at least. :p

18:49 < TD> You would be a legend

18:52 < rcurtin> if either of you guys would like to write up or add some documentation it would be greatly appreciated

18:53 < rcurtin> I personally don't use Windows so I have never written good instructions for how to do it :(

18:57 < TD> I don't mind adding anything I can

19:04 < nilay> zoq tham: hello again, so here is the code for feature extraction. https://gist.github.com/nilayjain/6c426cd4b02f3c4174dc892057311f1d

19:06 < nilay> so right now i am not using templates, even though class is templatized.

19:07 < nilay> what should i do in order to do that. also the varaibles that are being assigned in the code to matrices should also be of that same type, so thats why i am unsure.

19:10 < nilay> zoq: it takes 18s on my machine just for the example code that you posted. the slowest step for python codes is the reshape() that is being done in reg_ftr and ss_ftr

19:13 < nilay> i think even though we avoid that thing upto some extent, it may be slowest step for this implementation as well. So once the template thing is done, i can write it in mlpack format and we can use github then.

20:01 < zoq> nilay: So, you can basically use MatType instead of arma::mat in your code at the postions where it makes sense. So, instead of: void Gradient(arma::cube& InImage, arma::mat& Magnitude, arma::mat& Orientation);

20:01 < zoq> write: void Gradient(CubeType& InImage, MatType& Magnitude, MatType& Orientation);

20:01 < zoq> nilay: But it doesn't make sense to use arma::mat (arma::Mat<double>) to store the locations, so you can just use umat or arma::Mat<size_t>

20:02 < zoq> nilay: We take a look at the code, once you opened the PR, it's easier to start a discussi and to make comments.

20:03 < zoq> nilay: You can also take a look at e.g. https://github.com/mlpack/mlpack/blob/master/src/mlpack/methods/naive_bayes/naive_bayes_classifier.hpp how to use MatType

20:06 < zoq> nilay: Also if you open a PR, I think using 'edge_boxes' as method/folder name is a good idea. What do you think?

20:08 < nilay> zoq: yes, ok

20:09 < nilay> so, now mlpack needs to be recompiled fully, when i add files to cmakelists

20:10 < keonkim> hmm thats strange.

20:10 < keonkim> the master branch on github works but the 2.0.1 release fails cmake on cli. (with same armadillo and everything)

20:11 < keonkim> it says C++ 11 compiler is not available

20:11 < keonkim> I am talking about windows cmake cli.

20:12 < rcurtin> yes, I think for visual studio 2.0.1 has a bug... take a look at CMake/CXX11.cmake

20:14 < zoq> nilay: Just run 'make' it should automatically reconfigure your project.

20:15 < nilay> zoq: what about the internal variables that need to be created sometimes, they should also be of type MatType

20:15 < nilay> how do we ensure that?

20:15 < nilay> say when these variables will be assigned to a location in a MatType matrix.

20:19 < zoq> nilay: I'm not sure I get your point, you can also use MatType inside the function.

20:20 < nilay> zoq: say a matrix m is of type MatType

20:20 < nilay> now to matrix m at location (i,j) i need to assign a variable.

20:21 < nilay> so the type of that variable(float or double) should match with type of matrix(float or double)

20:22 < zoq> nilay: you mean something like arma::mat input = arma::mat(10, 10); input(0, 0) = 1?

20:22 < nilay> yeah

20:23 < nilay> is there a way to do this

20:24 < zoq> nilay: It should be casted automatically, we don't have to worry about that.

20:24 < nilay> zoq: ok

20:25 < nilay> so i'll open PR in the morning. then we'll discuss.

20:26 < zoq> nilay: sounds, great

20:26 < nilay> meanwhile should i start reading up on the next thing we want to do, when you and tham review this?

20:28 < zoq> nilay: That sounds like a great idea. Also, I'm stoked to read the next status update :)

20:30 < nilay> zoq: ok, i'll post that tomorrow too :) so we will be using decision_stump for the next part?

20:31 < zoq> nilay: Yeah, I think that's a good idea.

20:32 < nilay> zoq: ok, and we'll use python codes as reference for training random forest?

20:33 < zoq> nilay: We don't have too, we can just reuse ryan's code: https://github.com/rcurtin/mlpack/tree/vfdt/src/mlpack/methods/hoeffding_trees

20:34 < nilay> zoq: this seems like too many files :p

20:35 < zoq> nilay: We don't need everything, you could start with hoeffding_forest_impl.hpp and hoeffding_forest.hpp

20:38 < nilay> zoq: ok great. now i have idea how to start for the next thing. see you tomorrow :)

20:39 < zoq> nilay: See you tomorrow ... also have fun.

20:41 < keonkim> TD: I wrote down the instructions of "how to build mlpack on windows" on my blog : http://keon.io/mlpack-on-windows.html

20:42 < keonkim> TD: it is little rough since i rushed it :), also, try the one on the master branch (https://github.com/mlpack/mlpack) instead of the 2.0.1 release.

20:44 < zoq> keonkim: I don't mind to use either parameter.empty() or CLI::HasParam(); also I'm not sure we have to unify the executables, do you think it would increase readability?

20:46 < keonkim> zoq: yes I suggested it just for the readability.

20:51 < TD> Keonkim: Love the screen shots and thanks for putting this together! This explains a lot. I will send over any documentation that I can add

20:52 < zoq> keonkim: Okay, I don't see a reason not to do so, maybe some else does, but if not I would say go ahead :)

20:56 < zoq> keonkim: I agree, great blog post; step 3 lacks the armadillo build step, but I'm sure anybody can figure it out.

20:58 < keonkim> I just updated it. :) thanks for letting me know.

21:04 < keonkim> TD: tell me if you get unsuccessful with this instruction. I am going to sleep now. :)

21:06 < TD> Thanks and get some rest. I am going to give it a go tomorrow and I will let you know.