05:33 < jenkins-mlpack> Project mlpack - nightly matrix build build #449: STILL UNSTABLE in 1 hr 33 min: http://big.cc.gt.atl.ga.us:8080/job/mlpack%20-%20nightly%20matrix%20build/449/

09:00 < ryde> ..

09:00 < ryde> from /local/mhhakala/mlpack/mlpack-1.0.8/src/mlpack/core/dists/discrete_distribution.cpp:22:

09:01 < ryde> /share/apps/armadillo/4.100.0-mkl/usr/local/include/armadillo:59:27: error: mkl_service.h: No such file or directory

09:02 < ryde> so it refers to armadillo actually

09:03 < ryde> lots of similar issues with different mkl-related headers and functions

10:52 < ryde> well, now I got a bit further. I was using intel-c-compiler in the previous step. I dropped that and went on using linux gcc (while still using mkl libs) for armadillo.

10:52 < ryde> Still, the compilation fails as linker does not know about MKL

10:52 < ryde> [ 29%] Built target mlpack

10:52 < ryde> Linking CXX executable ../../../../bin/sparse_coding

10:52 < ryde> CMakeFiles/sparse_coding.dir/sparse_coding_main.cpp.o: In function `void arma::gemv<false, false, false>::apply_blas_type<double, arma::Mat<double> >(double*, arma::Mat<double> const&, double const*, double, double)':

10:53 < ryde> sparse_coding_main.cpp:(.text._ZN4arma4gemvILb0ELb0ELb0EE15apply_blas_typeIdNS_3MatIdEEEEvPT_RKT0_PKS5_S5_S5_[void arma::gemv<false, false, false>::apply_blas_type<double, arma::Mat<double> >(double*, arma::Mat<double> const&, double const*, double, double)]+0x191): undefined reference to `wrapper_dgemv_'

10:53 < ryde> how should I tell MLPACK

10:53 < ryde> 's cmake where to look for those?

12:32 < ryde> ok, reading a bit more on discussion list gave the help I needed. MLPack is now installed with Armadillo+MKL. Thanks for all the help.

14:27 < oldbeardo> naywhayare: is there a way to take argmax in armadillo?

14:30 < naywhayare> there is an overload of max() that returns the index

14:32 < oldbeardo> yes, but that is just for one element, what if I want the column wise arg max of a matrix

14:32 < naywhayare> I don't think there is a good way to do that; you might have to write it by hand

14:32 < naywhayare> Conrad might accept a contribution into Armadillo if you wrote up another overload of max() that did a column-wise argmax or something like that

14:34 < oldbeardo> okay

16:23 < oldbeardo> naywhayare: good news, the implementation is working

16:23 < oldbeardo> gives an accuracy of 79% when trained with 1000 examples from MNIST

16:26 < oldbeardo> I tested with 1000 examples