Ryan Curtin

Yes, I'd keep it pretty coarse-grained: basically mlpack failures will have one of two reasons: the data is screwed up or the algorithm failed to converge. Honestly the `std` exceptions...

Like I said, I don't see the need for an mlpack exceptions base class, since any of the problems we are encountering can be described as `invalid_argument` or `runtime_error` or...

Yes, is essence, that is the idea. The backtrace system Kirizaki wrote might be a good start at an idea but we need something that can work even without debugging...

There is only a need to pass-by-value in instances where the method actually modifies the data. So if we have a method `Predict()` which doesn't need to modify the input...

I'm not sure I understand the difference. There are two possibilities: - If the model can do predictions without modifying the data, we provide only `Predict(const mat&)`. - If the...

Yes, go ahead---but I would suggest doing small pieces at a time instead of the entire thing at once, to make reviewing easier.

Yes, it is (it would be closed if it was done), but like I wrote in a previous response, I'd suggest doing small pieces at a time. Feel free to...

Hi @hassan-mahmood, that is correct. If you'd like to work on this and open a PR, I would be happy to review and merge it. Thanks!

Hi @anky2936, this is a good issue to start on. Try to follow the style guide with your modifications, and be sure to build the code and run the tests...

@matiu2: the only issue with passing a regular reference instead of an rvalue reference is that there may be situations where we want to pass a temporary matrix to the...