05:29 < dms> hello guyz can you tell me how to know about various folders in mlpack-2.0.2 ?

05:30 < dms> What they do ? And how can they be used ? Should I know about all ? Or is it not necessary to do that ?

05:31 < mentekid> dms: do you mean inside src/mlpack?

05:46 < dms> yeah

05:47 < mentekid> well, I'm not sure, but I would say

05:48 < mentekid> core/ has the main functionality, that's where stuff like loading data/ what headers are included is

05:48 < mentekid> it also has stuff that is common to many algorithms, like distance computations, kernels, tree traversal etc

05:49 < dms> okay ..and what about folders outside src ?

05:49 < dms> do we need to care them ?

05:49 < mentekid> methods/ implements the machine learning methods mlpack offers - for example in methods/pca there is code for Principal Component Analysis. In each folder there's source code for an execuatble (I think!) that ends up in build/bin/ after compilation

05:50 < mentekid> depends what you need to do - CMake is files for the cmake program (which configures how mlpack is going to be built)

05:51 < mentekid> doc has the documentation, so for example if you can't access mlpack.org/docs.html for some reason you can find it there

05:51 < mentekid> but the code is in src/mlpack

05:52 < dms> ohkay ..so tell me how can I implement any algorithm with any command?

05:52 < mentekid> you mean a new algorithm?

05:53 < dms> what should i type in my terminal to implement some learning algorithm like linear regression

05:53 < mentekid> for mlpack? Or just use mlpack to implement your own stuff?

05:53 < mentekid> ah

05:53 < dms> use mlpack to implement my sutff

05:54 < mentekid> There's two ways. You can use the binaries in build/bin, for example it would be build/bin/mlpack_linear_regression or something for LR

05:54 < mentekid> (I'm not sure how the binary is named, I haven't compiled it because I don't need it)

05:55 < dms> how to load data in terminal ?

05:55 < mentekid> so you type bin/<binaryname> -h which will give you the help on how to use it. Each binary has ways to load training/testing data

05:56 < dms> okk and ?

05:56 < mentekid> give me a moment

05:58 < dms> sure

06:00 < mentekid> so, yeah, you look into the help output to find what switches you need (switches are the -<letter> or --<phrase-connected-by-dashes> commands)

06:01 < mentekid> so for example you might need to specify the output file, you find what the switch is for that, let's say -o, and you type bin<binaryname> -o myoutput.csv

06:02 < mentekid> you add as many switches and their arguments (if any) as you want, and in any order (there are some combinations that don't make sense, mlpack will produce an error to let you know)

06:02 < mentekid> for example, I have the iris_train.csv and iris_test.csv and I want, for every point in iris_test.csv to find the 5 nearest neighbors and store their IDs in iris_5nn.csv

06:02 < mentekid> I would type

06:03 < mentekid> bin/mlpack_knn -k 5 -r iris_train.csv -q iris_test.csv -n iris_5nn.csv

06:04 < dms> oh thatz good ..thank you!

06:05 < mentekid> no problem happy to help :)

14:31 < travis-ci> mlpack/mlpack#1042 (master - 2bd1227 : Ryan Curtin): The build passed.

14:31 < travis-ci> Change view : https://github.com/mlpack/mlpack/compare/ebf0dfcf067b...2bd1227d0f41

14:31 < travis-ci> Build details : https://travis-ci.org/mlpack/mlpack/builds/139208267

17:33 < marcosirc> @zoq

17:34 < marcosirc> hi, how are you?

17:34 < marcosirc> I have been working on the benchmarking system. I added a new "view" as you suggested.

17:36 < marcosirc> I was wondering why timing is not considered as the metric. This would make the code simpler, specially for the views.

17:36 < marcosirc> *as a metric.

17:43 < zoq> marcosirc: hm, at the time we implemented the timing part we didn't think about adding other metrics, but now that you brought it up, it sounds like a good idea. It should be straightforward to merge the timing and metric results.

17:45 < marcosirc> Yeah. I think so!