01:46 < stephent1> naywhayare: is there any way to get more info from http://big.cc.gt.atl.ga.us/job/mlpack%20-%20nightly%20matrix%20build/arch=linux-amd64,armadillo=armadillo-4.500.0,buildmode=release/730/console

01:46 < stephent1> like which specific test it was?

01:52 < stephent1> i did this

01:52 < stephent1> for i in {1..1000}; do ./bin/mlpack_test --run_test=SdpPrimalDualTest/SmallMaxCutSdp >> logx; done

01:52 < stephent1> and got no errors

01:53 < stephent1> i'm trying all of them now

01:59 < stephent1> naywhayare: does every new invocation of mlpack_test use a different seed?

02:10 < naywhayare> stephent1: no, it doesn't by default

02:10 < naywhayare> for the sake of reproducibility on a single system

02:11 < naywhayare> your best bet is to use math::RandomSeed(std::time(NULL)), and print the seed, then reuse a seed that causes it to fail

02:11 < naywhayare> you can get more information from the Jenkins build here:

02:11 < naywhayare> http://big.cc.gt.atl.ga.us/job/mlpack%20-%20nightly%20matrix%20build/arch=linux-amd64,armadillo=armadillo-4.500.0,buildmode=release/ws/reports/tests/

02:12 < naywhayare> (i.e. looking in the workspace, then in tests/; the mlpack_test.xml file should print all the tests that ran before failure)

02:12 < naywhayare> but in this case the workspace doesn't report failure, because this configuration worked this time, and the workspace is only the most recent build

02:12 < naywhayare> http://big.cc.gt.atl.ga.us/job/mlpack%20-%20nightly%20matrix%20build/arch=linux-amd64,armadillo=armadillo-4.450.4,buildmode=release/ws/reports/tests/ a better example, maybe

02:13 < naywhayare> I see this at the bottom of mlpack_test.xml:

02:13 < naywhayare> <TestCase name="SmallMaxCutSdp">

02:13 < naywhayare> error: chol(): failed to converge

02:13 < naywhayare> (then EOF, because Log::Fatal was called and that logs to stderr not stdout)

04:08 < stephent1> ok got it

04:08 < stephent1> i'll try this later tohnight

09:01 < stephent1> naywhayare: the bug is a bit subtle

09:01 < stephent1> i'm able to reproduce it

09:01 < stephent1> not quite sure how to fix it yet haha

09:09 < stephent1> naywhayare: alright i used a more stable starting point for now

09:09 < stephent1> https://github.com/mlpack/mlpack/commit/e4330544cb04ac6d2bb38e1372bffd8c6cef6bcb

09:09 < stephent1> i think w/o implementing preconditioning it may be hard to guarantee that all starting points will dos omething reasonable

12:49 < vlad_gl> Hi all! I want try to implement in place matrix transpose for #203. Where should i add this method?

13:26 < vlad_gl> and what is the correct way to run mlpack_test? if i run it from bin directory, it can't find test sets. need to specify the directory path?

13:44 < naywhayare> vlad_gl: when you 'make mlpack_test', it automatically moves all the test sets to the build/ directory

13:44 < naywhayare> so you can run it from build/... 'bin/mlpack_test'

13:45 < naywhayare> you can run individual test suites too, i.e. bin/mlpack_test -t LogisticRegressionTest (or whatever suite; you can find the names of the suites and of individual tests in src/mlpack/tests/)

13:45 < naywhayare> if you implement the in-place matrix transpose, it may be suitable for contribution to armadillo, or maybe we should keep it in arma_extend/

13:46 < naywhayare> so, write it, let's see how it performs, and if it doesn't cost a huge amount of extra runtime (or it's faster), we can find a way to work it in

13:49 < vlad_gl> yeah, i understand. i run mlpack_test then i was in bin directory using ./mlpack_test, not in build directory and ./bin/mlpack_test

13:49 < vlad_gl> thanks

14:11 < vlad_gl> armadillo alredy have 2 implementations of inplace trans methods. heh.

14:11 < vlad_gl> have you tried them?

14:20 < naywhayare> vlad_gl: hmm, maybe this will be much simpler than I thought then :)

14:20 < naywhayare> I think at the time that ticket was written, maybe the inplace transpose wasn't there yet...

14:22 < naywhayare> the right way to test it would probably be to use it in data::Load() (src/mlpack/core/data/load_impl.hpp, I think) and see how it compares (w.r.t. runtime) for various datasets from small (a few kb) to large (500MB+)