13:22 < MattGomes> Does anyone recommend a particular area of the codebase to start understanding the core of mlpack? I have been playing with the Python bindings and seeing where they map to in the c++ code, so I was wondering if anyone recommends a different approach

14:27 < extrowerk> Hi

14:28 < extrowerk> So somebody recommended me many times to enable the python wrapper in my mlpack Haiku port.

14:29 < extrowerk> I haven't did that before, because the installer puts the python stuff in a wrong folder where python can't find it.

14:31 < extrowerk> But now i found out where to move those files, and it seems to be working (at least with opencv: http://0x0.st/iRFc.png ) so i expect this trick should get the mlpack python wrapper working aswell.

14:31 < extrowerk> Will patch our recipe

16:06 < zoq> rcurtin: Trying to integrate the cooperative groups based accu kernel, but I'm getting "error: name followed by "::" must be a class or namespace name" when I try to use cooperative_groups::this_thread_block();. I included #include <cooperative_groups.h> in include/bandicoot, which should be all that is needed, have you seen a similar issue before?

16:08 < NippunSharmaGitt> Hi, I have a doubt [here](https://github.com/mlpack/mlpack/blob/49f13e59682d4ae12882a74d6d58efb05540f0f4/src/mlpack/methods/ann/layer/recurrent.hpp#L5) . Shouldn't this comment state definition of `Recurrent` class and not `LinearLayer` ?

17:01 < rcurtin> zoq: no, I haven't seen anything like that, maybe the documentation is wrong or something?

17:02 < rcurtin> (like maybe the name of this_thread_block() is actually different?)

17:02 < rcurtin> is this inside of the kernel code, or on the host?

17:02 < rcurtin> also, I rebuilt the website and it looks like the IRC logs are working again

17:03 < zoq> rcurtin: The code works outside of bandicoot, just using nvcc, so I guess the documentation is correct..

17:03 < zoq> rcurtin: Also it's inside the kernel code.

17:03 < rcurtin> oh, strange... this is a CUDA kernel, right?

17:04 < zoq> correct

17:04 < rcurtin> all the CUDA kernels are compiled with NVRTC, not with nvcc, so I think it might be possible that there is some option there that's set wrong or something?

17:04 < rcurtin> take a look at cuda::runtime_t::compile_kernels() in include/bandicoot_bits/cuda/runtime_meat.hpp

17:04 < zoq> Okay, that is a good start, I'll see if I can find anything.

17:05 < rcurtin> maybe it needs another option in `opts`?

17:06 < zoq> Maybe, I already had to use another gpu-architecture, since compute_30 isn't supported with cuda 11

17:07 < zoq> NippunSharmaGitt: Nice catch, you are right.

17:08 < rcurtin> yeah, I hardcoded the GPU architecture but almost certainly that should be changed

17:10 < zoq> Ahh, nice there is an option to pass headers, that should solve the issue, thanks.

17:11 < rcurtin> sweet