10:38 < zoq> rcurtin: Great, once I solved the python build issue on windows, I can create a package for windows.

13:15 < travis-ci> Prabhat-IIT/mlpack#1 (master - b288571 : Ryan Curtin): The build has errored.

13:15 < travis-ci> Change view : https://github.com/Prabhat-IIT/mlpack/compare/9e108640748b2a78bde60097b32dd808d9036a82...b288571029f30a1ab128c8fb9fe77727ac28a85e

13:15 < travis-ci> Build details : https://travis-ci.org/Prabhat-IIT/mlpack/builds/363485064

16:25 < travis-ci> Prabhat-IIT/mlpack#2 (fastmksTest - 90970c9 : Prabhat): The build has errored.

16:25 < travis-ci> Change view : https://github.com/Prabhat-IIT/mlpack/compare/b60f4443ae49...90970c9c3cb1

16:25 < travis-ci> Build details : https://travis-ci.org/Prabhat-IIT/mlpack/builds/363519406

16:37 < manthan> zoq : with respect to your comment on flexibleReLU PR, due to the fact that since input is zeroing out and so pertubations won't cause any change, the gradient should be the same with and without pertubations and so test should pass right?

16:47 < manthan> zoq : what i mean is i understood why is estGradient evaluating to 0 but this same thing should be valid when the whole module is run right?

17:23 < zoq> The complete gradient step uses the alpha parameter as output, but the gradient check is based on the assumption, that a permutation in one direction has an effect on the result, which isn't the case, in the range of 0 and < 0.

17:28 < manthan> ya agreed with that but i don't understand why was it failing in specific test check and passing in complete module test

17:28 < manthan> also the modified test looks good and i will update the PR now :D

17:34 < zoq> On my system both cases fail, my guess is the complete run does use another random seed where the weights are positive and the single run doesn't.

17:34 < manthan> ya even i almost concluded that :D

17:34 < manthan> thanks

21:10 < xa0> I'm seeing a fatal error (munmap_chunk(): invalid pointer) when I try and use softmax regression

21:10 < xa0> https://i.imgur.com/bjycubV.png

21:17 < rcurtin> xa0: munmap_chunk()... are you using mmap() to allocate a matrix?

21:17 < xa0> nope. have a look at the backtrace, the error comes from mlpack

21:17 < xa0> specifically, the destructor of softmaxregressionfunction

21:17 < xa0> i only ever use the standard constructors for matrices

21:19 < rcurtin> ah, well I am greasy right now (I am working on a car) so I can't look into it at the moment, but if you want to try and make a minimum reproducible example I can look into it when I have a chance

21:19 < rcurtin> it does seem very strange to me that munmap_chunk() is being called, I suspect something weird is going on

21:19 < xa0> sure ^^

21:21 < rcurtin> you can file a github issue too, maybe someone not on IRC might be able to reproduce before I can get around to looking into it

21:23 < xa0> ah okay, will do

22:14 < xa0> issue created: https://github.com/mlpack/mlpack/issues/1358

23:48 < xa0> i've used valgrind and found some uninitialized values, but as far as i can tell, and i really hope i'm wrong cause i don't want to have to upgrade to the git version, but i'm pretty sure they're not coming from me