This fixes #252 and also the corresponding issue downstream.
I checked that the performance (both AD and forward) is pretty much the same (slightly better in my setup) as the previous implementation. Interestingly, doing mapreduce(f, vcat) on ys performed consistently worse than doing map+reduce(vcat) and also the previous implementation.
This fixes #252 and also the corresponding issue downstream.
I checked that the performance (both AD and forward) is pretty much the same (slightly better in my setup) as the previous implementation. Interestingly, doing
mapreduce(f, vcat)
onys
performed consistently worse than doingmap
+reduce(vcat)
and also the previous implementation.