07:04 < billLiu> I use MKL to compile Armadillo but when I train GMM, I got this messege : "Intel MKL ERROR: Parameter 8 was incorrect on entry to DGEMM". Any idea about what I did wrong?

07:04 < billLiu> Then it says "Segmentation fault (core dumped)"

16:34 < billLiu> OK, I fix that problem by using OpenBlas instead of MKL, it works. I don't know what is the reason for that...

16:37 < billLiu> naywhayare, I scan the GMM training code and found the most time consuming part is initial clustering. When the cluster is done, multiprocess matrix processing will speed up training. And I add some simple openMP code to parallelize KMeans. Can you tell me how I could compile mlpack with openMP? Further, if I wanna update my version, where is the way?

23:26 < naywhayare> billLiu: look for the snippet of code in CMakeLists.txt with the comment "We require OpenMP now." and uncomment that code

23:27 < naywhayare> then, when you build mlpack, it'll automatically pass -fopenmp to the compiler (or whatever the correct flags are)

23:27 < naywhayare> it seems odd to me that the initial clustering is taking the most time. I think it does several runs of this clustering; maybe parallelizing that is a good solution?

23:28 < naywhayare> to keep your code up to date, you can make your modifications to a branch of the github repo ( https://github.com/mlpack/mlpack ) and 'git pull' occasionally to pull in any changes that have been committed to the master branch