Kyle Daruwalla
Kyle Daruwalla
> Batches don't have to be in a "sequence" to be fed into the model but a batch should have sequences. Quite the opposite for Flux as Brian pointed out....
Yeah I like this approach better because of the unification. It addresses the concerns about tying `batchseq` into the data block visualization. Now, it should be clear to the user...
It might be worth also looking at [DataSets.jl](https://juliahub.com/docs/DataSets/4adr3/0.2.4/) announced at JuliaCon.
> Many distributions are already generic. It would be great if you could prepare PRs for the ones that aren't. Yeah I'll start looking into this. > Can't you use...
I got pulled into other projects that prevents me from focusing on this issue right now. But I know @srinjoyganguly is working on a DistributionsGPU.jl package.
I opened a new issue for GPU-related support ( #1067 ) to coalesce the steps it would take to get the package there. Happy to start contributing towards this goal...
Yes, NNlib should get its own documentation.
I think having an abstract type that's used internally to remove duplication is fine (though I think you really only remove the forward pass and maybe `@functor`?). What I want...
How are you trying to do it? By using the macro? That won't work. You need to define the function `functor` for the abstract type.
I totally agree that we should have these built-in pipelines, but I don't think Metalhead.jl is the right place for them. I would suggest MLDatasets.jl instead, since generally the pre-processing...