Ryan Curtin

> Yes this would require a change to CARMA but if really needed we could maintain a separate branch that uses the default Armadillo allocators and copies when on windows....

Sounds good! Our Slack room is bridged to our Matrix room and Gitter room (but currently not IRC ...), so you can use any of those protocols without a problem....

@RUrlus thanks for the update---please don't hesitate to ask if you'd like me to explain anything about the binding generator setup. It would be exciting to transition away from Cython...

I think, at least for the MNIST example here, part of the reason is that each decision stump can only split two classes (at most) on one feature, so decision...

No need to assign it, just work on it and open a PR. :+1: This PR will require a decent bit of investigation about how our defaults perform on different...

@dnabanita7 the code snippets may be outdated, but the underlying issue should still be relevant. Can you post more information about the problems you are having?

You don't need to use the hyperparameter tuning or anything. You can just use the command-line program or the Python bindings to replicate the original finding that convergence is not...

Two updates relevant to this issue: - #2370 contains some nice utilities for checking the size of a matrix. We should probably use those in whatever we build here. -...

@VedangAsgaonkar yes, as you can see, the issue is not closed, so it is still open. :smile: Please feel free to take a look at some of the PRs that...

> @rcurtin I thought more about the "include everything in one header" issue (follow-up to #3233). Rather than making the "one header" approach mandatory and converting all of the codebase,...