Anthony Blaom, PhD comments

Results 815 comments of


                                            Anthony Blaom, PhD

mlj_model macro creates methods with bad line number info

Many thanks for the expert feedback.

Should a history of losses and training losses be explicity returned?

I think a better (more general) idea is to add a wrapper that replaces the returned report for a control with a complete history of state for that control.

Perhaps MLJModelInteface should not export any method in the future

Or, at least it should not export any "dummy" method that gets re-defined by MLJBase.

Support ProgressMeter or other progress logging utiliites such Term.jl?

Thanks @liuyxpp for this observation. Refactoring current logging to `Info` in this way sounds like a good idea. PR welcome.

It seems some macros in metadata_utils cause issues if an MLJ interface is wrapped in module

I wonder if https://github.com/JuliaAI/MLJModelInterface.jl/pull/184#issuecomment-1864104215 fixes this issue?

Improve local testability

Sure, unless you have a better suggestion.

Improve local testability

Helpful comments, @deyandyankov, thank you. > Another way to go would be to mock the mlflow endpoints. This will not require an mlflow instance at all. That's probably the correct...

Improve local testability

This sounds like a nice idea!

Adding `use_ajax` flag to `MLFlow` instances

No objections but this definitely needs to be tagged breaking. So if we have any other breaking changes in mind...

Integrate with DagsHub's MLflow API

Thanks @pebeto for taking the lead here. Your suggestion sounds reasonable to me, but would be good to get @deyandyankov's view. Are you adding `auth` as a new keyword but...