Mike J Innes
Mike J Innes
That makes sense. Should be fairly easy to add a check for this.
Thanks for the report. We should really add some logging so that we can see what patch failed where; you might be interested to add some logging to [this line](https://github.com/MikeInnes/Vimes.jl/blob/310c4ac574ebe1f3cc186b794f0ac6a80225a4fa/src/patch.jl#L10)...
This seems like a good idea, but I'm not sure how to present it to users either. Diffs are not ideal, since for example a mutation that replaces an `Int`...
It's possible that we'll just end up having to fork CSTParser for stuff like this. Not ideal but what we want is really quite different; it'd be nice to have...
I used to use `cu` in Flux but ended up going against that; `adapt` means something different and it doesn't really work to apply it to models. It might be...
Sure, I'm not expecting you to take a Flux dependency. Splitting out the `treelike` stuff would be the right move. The reason for `gpu` as opposed to `cu`, in Flux,...
For Flux's purposes at least, we don't really need AbstractTrees' functionality. We really just need `mapchildren` and then the rest follow from there.
How about turning this into a Julia package? Makes the dependencies easier. It would be nice to start training a standard model in a couple of lines, and even better...
I'm a bit wary of exposing `@nograd` because it tends to be overzealous (e.g. adding it to `rand` causes issues). Would a version that requires a signature be reasonable? Not...
Yeah, I can give it a shot at some point.