Kyle Daruwalla

One of the things I would like to determine is how much of the non-getobs/nobs code can be removed (or at least is there value in keeping it). If there...

Yeah basically if there is an abstract type like `SlidingWindow` that actually gets used (to define some shared code), then keep it. If there is an abstract type that is...

@racinmat had to hand back off to me due to work commitments. I have been busy all semester, but I will work on this during break.

Haven't tested this, but this a rough sketch of the approach. Probably needs more iterations in the oven, but the more we take Functors.jl in this direction, the better. Even...

The major breakages will be w.r.t. Flux's uses of Functors. There are probably few users directly using Functors outside of `@functor`. So, if we need to scrap one day, then...

This is a bit of a drive-by comment, some of this might already be addressed. See some earlier discussion in #21. I think in general this would make sense to...

> Transpose, Adjoint, and maybe ReshapedArray, PermutedDimsArray. For the reason that shared weights will often be transposed. Depending on the shared weight solution, these types might be ones we don't...

For this feature, it would be nice to have it as an option when you open a tab space. So instead of always closing the current tabs and opening the...

Not state, but a hyper-parameter that belongs to the training loop and not a callback. I basically have a setup where each batch is time series that computes a loss...

Ah I didn't think of that. At least for my case, adding a field to the phase will be much more intuitive, so I probably won't use this feature. I'll...