do not differentiate through the construction of BitArray #506
Conversation
|
Cool, LGTM. Sometimes we test them, if we have had them defined wrong in the past. |
|
Will merge once CI passes. |
|
cool, Thanks a lot! |
| @@ -84,6 +84,7 @@ | |||
| @non_differentiable Vector(::AbstractArray{Bool}) | |||
|
|
|||
| @non_differentiable Array(::AbstractArray{Bool}) | |||
| @non_differentiable BitArray(::Any) | |||
There was a problem hiding this comment.
Is there a reason we didn't splat this?
@non_differentiable BitArray(::Any...)
There was a problem hiding this comment.
Doesn't it only have 1 constructor?
I didn't actually check
There was a problem hiding this comment.
julia> methods(BitArray)
# 6 methods for type constructor:
[1] BitArray(::UndefInitializer, dims::Integer...) in Base at bitarray.jl:69
[2] BitArray(::UndefInitializer, dims::Tuple{Vararg{Integer, N}}) where N in Base at bitarray.jl:71
[3] BitArray(x::BitArray) in Base at bitarray.jl:543
[4] BitArray(A::AbstractArray{var"#s79", N} where var"#s79") where N in Base at bitarray.jl:503
[5] (::Type{T})(x::T) where T<:BitArray in Base at bitarray.jl:542
[6] BitArray(itr) in Base at bitarray.jl:574there seems to be more
There was a problem hiding this comment.
Should I make another PR?
There was a problem hiding this comment.
Please do, that would be great, thanks!
Based on @oxinabox 's reply here: SciML/DiffEqFlux.jl#502 (comment) .
This solves also the issues with the callback construction in SciML/DiffEqFlux.jl#502.
Does it need a test? Should I maybe add one from SciML/DiffEqFlux.jl#502 (comment) in here: https://github.com/JuliaDiff/ChainRules.jl/blob/master/test/rulesets/Base/array.jl ?