Anthony Blaom, PhD

icon indicating a website link ablaom.github.io icon indicating an email [email protected]

icon company University of Auckland icon location Auckland icon location Mathematician and Data Scientist, PhD Caltech • Expertise: Differential Geometry, Dynamical Systems, Fluid Mechanics, Statistics, Machine Learning, Julia

Results 815 comments of Anthony Blaom, PhD

Add `total(f, model)` to replace implicit `sum(f, Flux.params(model))`

Thanks for the lightning reply ❤️

CompatHelper not working

@cscherrer Could you try renewing your keys to see if that clears this up?

MLUtils seems quite heavy

Anyone have some time to revisit this?

MLUtils seems quite heavy

Related (duplication?): https://github.com/JuliaML/MLUtils.jl/issues/90

Archiving DataLoaders.jl

> As a further breakdown, I also suggest to move getobs and numobs and their implementations for the Base types to a MLCore.jl repo. Yes, please.

`params(::MLJType)` method

In work on MLFlow integration we see that we need such a function. So opening a new issue to add a `params2` which in the next breaking release replaces `params`...

Explain in docs that model wrappers need to overload save/restore

And forward other methods/traits, such as `feature_importances`, and so forth (see BinaryThresholdPredictor code for other examples)

Explain in docs that model wrappers need to overload save/restore

Also, after https://github.com/JuliaAI/MLJ.jl/issues/1125 is closed, they will need to overload `MMI.constructor`.

invalid model parameters should either error or keep user arguments

This makes sense. As far as changing the behaviour of the `@mlj_model` macro, that will have to wait for the next breaking release, which won't be for a while.

invalid model parameters should either error or keep user arguments

Thanks @ExpandingMan for these further thoughts. Sorry to hear about the parameter woes. Gradient boosters have a lot of them, that's for sure. In my experience maintaining the MLJ ecosystem,...