Kristoffer Carlsson
Kristoffer Carlsson
No, t can be a local variable and you send in the closure `x -> f(x, t)`. Just like always when you need to pass parameters.
Yeah but you can change `t` afterward. Perhaps that's what you meant.
Maybe, but I wonder if not just propagating `args...; kwargs...` (like you have in the PR I think) is easier in that case.
This should be possible, might just need to relax some of the parameters. I will look into it.
Wouldn't just computing knn by itself and then differentiating through the distance function give the same result?
It would be good to: 1. Give the input data distribution. Is it just random points? 2. Rewrite the code to use the NearestNeighbors.jl API.
I think you can do better by basically searching and pruning two trees against each other. I think I had an implementation of that but don't recall why I didn't...
Makes sense to have it here in my opinion.
Would be nice to run that benchmark here as well.
Hm, I didn't realize there was any thought of adding any binary dependencies to this repo.