00:58 < radens> mlpack::data::Split takes a matrix, a test ratio, and a tuple. http://mlpack.org/docs/mlpack-git/doxygen/namespacemlpack_1_1data.html#a80fb8838c66ff6351af3bbb33ee6ebe1

00:59 < radens> 1. If my test ratio is 0.7, which is the bigger element of the 2 tuple, the first element or the second?

00:59 < radens> 2. The split happens row-wise, and not column-wise, right?

00:59 < radens> I hope

15:30 < radens> rcurtin: Hey, I have a question about a type error with mlpack

15:30 < radens> Here is the error:

15:30 < radens> https://pastebin.osuosl.org/43796/

15:31 < radens> Here is the line of code throwing a fit: auto dstump = mlpack::decision_stump::DecisionStump<arma::Mat<int>>(data, labels, 2);

15:32 < radens> (data is an arda::Mat<int> and labels is an arma::Mat<size_t>)

15:32 < rcurtin> radens: it looks like DecisionStump isn't properly written so that it can handle any matrix type

15:33 < rcurtin> the problem appears to be in SetupSplitDimension... hang on

15:33 < rcurtin> also, to clarify something you mentioned last night, if the test ratio is 0.7, then 70% of the data will be in the test set

15:34 < radens> Thanks

15:34 < rcurtin> and also the split will happen column-wise---each point in mlpack is a column, not a row; see http://mlpack.org/docs/mlpack-2.1.0/doxygen.php?doc=matrices.html for some more info

15:34 < rcurtin> basically, if you load a CSV with data::Load() or something like this, it's automatically transposed, so your rows in the CSV become columns in the arma::mat

15:35 < radens> Yeah, at first I thought that was going to be a problem, but then I realize that data::Load rotates the input of CSVs by default

15:35 < rcurtin> so the short solution to your current problem is to use arma::mat not arma::Mat<int>; the slightly longer solution is to wait a minute for me to get a patch together :)

15:35 < radens> Thanks!

15:35 < rcurtin> this should be a relatively simple fix, basically there is just code that assumes arma::mat when it should instead be using a template type

15:54 < rcurtin> radens: fixed in 9774978, pull from master and that should compile

15:57 < rcurtin> that was weird, irssi decided it was done accepting input

16:39 < travis-ci> mlpack/mlpack#1630 (master - 9774978 : Ryan Curtin): The build passed.

16:39 < travis-ci> Change view : https://github.com/mlpack/mlpack/compare/86689aa4f384...97749789344d

16:39 < travis-ci> Build details : https://travis-ci.org/mlpack/mlpack/builds/180437353

18:11 < radens> rcurtin: how can I deserialize an adaboost model?

18:11 < radens> Doing this doesn't seem to work: archive >> adaboost;

18:11 < rcurtin> radens: ah, here is a bit of trickiness... :)

18:11 < rcurtin> first the quick answer:

18:12 < radens> Are there examples somewhere? In the tests it does it with a name value pair.

18:12 < rcurtin> you can use data::Load() and data::Save() to serialize models, i.e.

18:12 < rcurtin> actually, we can do something simpler than that

18:12 < rcurtin> archive >> mlpack::data::CreateNVP(adaboost, "adaboost_model");

18:12 < rcurtin> in typical boost::serialization code, you would use BOOST_SERIALIZATION_NVP() not data::CreateNVP()

18:13 < rcurtin> but because mlpack style guidelines are that member functions start with capital letters, mlpack implements Serialize() functions not serialize() functions

18:13 < rcurtin> and then CreateNVP() is a complex template metaprogramming shim that makes it all work :)

18:13 < rcurtin> either way, you could also use data::Load() and data::Save(), which autodetect the type based on the extension, so something like

18:13 < radens> Let me try with data::{Load,Save} then

18:13 < rcurtin> data::Load("filename.xml", "adaboost_model", adaboost)

18:14 < rcurtin> data::Save({ the same parameters })

18:14 < rcurtin> those should work too

18:19 < radens> brilliant, thanks