05:17 < sshkhrnwbie> Hi mlpack devs

05:18 < sshkhrnwbie> I raised issue 1228 regarding missing implementations of Xavier and He initializations in the ann module

06:09 < rajiv_> Hello all! I was trying to debug a test for an optimizer... Is there a particular way to do it? I tried to compile the code using g++ from the terminal but I'm getting the error that the header files are not found. I change the path to absolute for some files and it worked but it is asking to change path in all the codes being compild which is not possible. So, how should I debig the code?

06:10 < rajiv_> *debug

06:22 < Harpreet> Hello!! I have built and installed mlpack 2.2.5 on ubuntu and trying out the command line utilitty by following tutorials. I facing an issue whenever I execute "mlpack_logistic_regression -t iris_train.csv -l iris_train_labels.csv" I get runtime error stating "[FATAL] The labels must be either 0 or 1, not 2!" Kindly help

06:44 < Harpreet> I guess mlpack_logistic_regression does not support multi-class logistic regression!!!

07:58 < sshkhrnwbie> I was thinking of implementing a variance scaling initializer ( which will use the already implemented gaussian and random initializers ) and then using it for both Xavier and He initialization as it has been done in Keras.

07:59 < sshkhrnwbie> I would like to have some feedback from the owners before I start implementing

11:28 < haritha1313> @harpreet: i think you can use mlpack_softmax_regression for multiclass classification

11:29 < haritha1313> this might be helpful http://mlpack.org/docs/mlpack-git/man/mlpack_softmax_regression.html

11:35 < Harpreet> Thanks

13:14 < zoq> rajiv_: What command did you use? Something like: g++ foo.cpp -o foo -std=c++11 /path/to/lib/libmlpack.so -I/path/to/mlpack/include/ -larmadillo -lboost_serialization -lboost_program_options should work.

13:15 < zoq> sshkhrnwbie: That makes sense to me.

13:18 < zoq> sshkhrnwbie: In case you missed the response: http://www.mlpack.org/irc/

13:19 < sshkhrnwbie> I had one more thing to ask

13:20 < sshkhrnwbie> In #939 Kris was implementing shapiro wilk test for testing the initializations. Ryan commented that there isn't a need for it. I am not familiar with this test and wanted to check if it will be required.

13:24 < zoq> There is no need to implement the test if you use GaussianDistribution as Ryan pointed out.

16:45 < sshkhrnwbie> Thanks for clarifying. I will go through your design guidelines tomorrow and submit a pr for the variance scaling initializer soon.

17:04 < zoq> sshkhrnwbie: Sounds good.