Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

destructure doesn't work correctly with certain functors #62

Open
rejuvyesh opened this issue Mar 23, 2022 · 3 comments · May be fixed by #70 or #63
Open

destructure doesn't work correctly with certain functors #62

rejuvyesh opened this issue Mar 23, 2022 · 3 comments · May be fixed by #70 or #63
Labels
bug Something isn't working

Comments

@rejuvyesh
Copy link

rejuvyesh commented Mar 23, 2022 •
edited
Loading 

using Flux
using Functors
using Optimisers

struct Custom
    abc::Tuple
end
Functors.@functor Custom (abc,)

function (f::Custom)(x)
    x .* f.abc[1] .+ f.abc[2]
end

function Custom(;dim::Int)
    abc = (randn(Float32, dim), randn(Float32, dim))
    return Custom(abc)
end

model = Flux.Chain(Dense(4, 16), Custom(;dim=16), Dense(16, 4))
p, re = Optimisers.destructure(model)
re(p)(randn(Float32, 4, 1))

leads to

ERROR: LoadError: type Tuple has no field layers
Stacktrace:
  [1] getproperty
    @ ./Base.jl:42 [inlined]
  [2] functor(#unused#::Type{Chain{Tuple{Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}, Custom, Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}}}}, c::Tuple{NamedTuple{(:weight, :bias, :σ), Tuple{Int64, Int64, Tuple{}}}, NamedTuple{(:abc,), Tuple{Tuple{Int64, Int64}}}, NamedTuple{(:weight, :bias, :σ), Tuple{Int64, Int64, Tuple{}}}})
    @ Flux ~/.julia/packages/Flux/qAdFM/src/layers/basic.jl:44
  [3] _trainable_biwalk(f::Function, x::Chain{Tuple{Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}, Custom, Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}}}, aux::Tuple{NamedTuple{(:weight, :bias, :σ), Tuple{Int64, Int64, Tuple{}}}, NamedTuple{(:abc,), Tuple{Tuple{Int64, Int64}}}, NamedTuple{(:weight, :bias, :σ), Tuple{Int64, Int64, Tuple{}}}})
    @ Optimisers ~/.julia/packages/Optimisers/UAVzc/src/destructure.jl:94
  [4] #fmap#30
    @ ~/.julia/packages/Functors/qBIlC/src/functor.jl:78 [inlined]
  [5] _rebuild(x::Chain{Tuple{Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}, Custom, Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}}}, off::Tuple{NamedTuple{(:weight, :bias, :σ), Tuple{Int64, Int64, Tuple{}}}, NamedTuple{(:abc,), Tuple{Tuple{Int64, Int64}}}, NamedTuple{(:weight, :bias, :σ), Tuple{Int64, Int64, Tuple{}}}}, flat::Vector{Float32}, len::Int64; walk::Function, kw::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ Optimisers ~/.julia/packages/Optimisers/UAVzc/src/destructure.jl:83
  [6] _rebuild
    @ ~/.julia/packages/Optimisers/UAVzc/src/destructure.jl:82 [inlined]
  [7] (::Optimisers.Restructure{Chain{Tuple{Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}, Custom, Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}}}, Tuple{NamedTuple{(:weight, :bias, :σ), Tuple{Int64, Int64, Tuple{}}}, NamedTuple{(:abc,), Tuple{Tuple{Int64, Int64}}}, NamedTuple{(:weight, :bias, :σ), Tuple{Int64, Int64, Tuple{}}}}})(flat::Vector{Float32})
    @ Optimisers ~/.julia/packages/Optimisers/UAVzc/src/destructure.jl:51
@mcabbott
Copy link
Member

Thanks, that's a bug.

It does work on Flux master BTW, as we changed Chain not to have children which aren't fields. But destructure ought to still allow such things.

julia> re(p)(randn(Float32, 4, 1))
4×1 Matrix{Float32}:
 -0.9730846
  4.3140717
  1.6813886
 -1.1657915

julia> Functors.functor(model)
((layers = (Dense(4 => 16), Custom((Float32[1.0382574, 2.225525, ...

julia> @which Functors.functor(typeof(model), model)  # the default
functor(::Type{<:Chain}, x) in Flux at /Users/me/.julia/packages/Functors/qBIlC/src/functor.jl:23

(jl_eaNYpA) pkg> st Flux
Status `/private/var/folders/yq/4p2zwd614y59gszh7y9ypyhh0000gn/T/jl_eaNYpA/Project.toml`
  [587475ba] Flux v0.13.0-DEV `~/.julia/dev/Flux`

@mcabbott mcabbott added the bug Something isn't working label Mar 23, 2022
@rejuvyesh rejuvyesh mentioned this issue Mar 23, 2022
Merged
@ToucheSir
Copy link
Member

What were you thinking for a solution for a quick release? Something like

functor(::Type{<:Chain}, c::Tuple) = c

?

@rejuvyesh
Copy link
Author

@ToucheSir that specific suggestion leads to:

ERROR: LoadError: type Int64 has no field weight
Stacktrace:
  [1] getproperty
    @ ./Base.jl:42 [inlined]
  [2] functor(#unused#::Type{Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}}, x::Int64)
    @ Flux ~/.julia/packages/Functors/qBIlC/src/functor.jl:23
  [3] _trainable_biwalk(f::Function, x::Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}, aux::Int64)
    @ Optimisers ~/.julia/packages/Optimisers/UAVzc/src/destructure.jl:94
  [4] #fmap#30
    @ ~/.julia/packages/Functors/qBIlC/src/functor.jl:78 [inlined]
  [5] #31
    @ ~/.julia/packages/Functors/qBIlC/src/functor.jl:78 [inlined]
  [6] #22
    @ ~/.julia/packages/Optimisers/UAVzc/src/destructure.jl:100 [inlined]
  [7] #4
    @ ./generator.jl:36 [inlined]
  [8] iterate
    @ ./generator.jl:47 [inlined]
  [9] collect(itr::Base.Generator{Base.Iterators.Zip{Tuple{Tuple{Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}, Custom, Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}}, Tuple{Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}, Custom, Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}}, NamedTuple{(:weight, :bias, :σ), Tuple{Int64, Int64, Tuple{}}}}}, Base.var"#4#5"{Optimisers.var"#22#23"{Functors.var"#31#32"{typeof(Optimisers.isnumeric), typeof(Optimisers._trainable_biwalk), IdDict{Any, Any}, Functors.NoKeyword, Optimisers.var"#20#21"{Vector{Float32}}}}}})
    @ Base ./array.jl:724
 [10] map(::Function, ::Tuple{Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}, Custom, Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}}, ::Tuple{Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}, Custom, Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}}, ::NamedTuple{(:weight, :bias, :σ), Tuple{Int64, Int64, Tuple{}}})
    @ Base ./abstractarray.jl:2948
 [11] _trainmap
    @ ~/.julia/packages/Optimisers/UAVzc/src/destructure.jl:99 [inlined]
 [12] _trainable_biwalk(f::Function, x::Chain{Tuple{Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}, Custom, Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}}}, aux::Tuple{NamedTuple{(:weight, :bias, :σ), Tuple{Int64, Int64, Tuple{}}}, NamedTuple{(:abc,), Tuple{Tuple{Int64, Int64}}}, NamedTuple{(:weight, :bias, :σ), Tuple{Int64, Int64, Tuple{}}}})
    @ Optimisers ~/.julia/packages/Optimisers/UAVzc/src/destructure.jl:95
 [13] fmap(f::Function, x::Chain{Tuple{Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}, Custom, Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}}}, ys::Tuple{NamedTuple{(:weight, :bias, :σ), Tuple{Int64, Int64, Tuple{}}}, NamedTuple{(:abc,), Tuple{Tuple{Int64, Int64}}}, NamedTuple{(:weight, :bias, :σ), Tuple{Int64, Int64, Tuple{}}}}; exclude::typeof(Optimisers.isnumeric), walk::Function, cache::IdDict{Any, Any}, prune::Functors.NoKeyword)
    @ Functors ~/.julia/packages/Functors/qBIlC/src/functor.jl:78
 [14] _rebuild(x::Chain{Tuple{Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}, Custom, Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}}}, off::Tuple{NamedTuple{(:weight, :bias, :σ), Tuple{Int64, Int64, Tuple{}}}, NamedTuple{(:abc,), Tuple{Tuple{Int64, Int64}}}, NamedTuple{(:weight, :bias, :σ), Tuple{Int64, Int64, Tuple{}}}}, flat::Vector{Float32}, len::Int64; walk::Function, kw::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ Optimisers ~/.julia/packages/Optimisers/UAVzc/src/destructure.jl:83
 [15] _rebuild
    @ ~/.julia/packages/Optimisers/UAVzc/src/destructure.jl:82 [inlined]
 [16] (::Optimisers.Restructure{Chain{Tuple{Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}, Custom, Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}}}, Tuple{NamedTuple{(:weight, :bias, :σ), Tuple{Int64, Int64, Tuple{}}}, NamedTuple{(:abc,), Tuple{Tuple{Int64, Int64}}}, NamedTuple{(:weight, :bias, :σ), Tuple{Int64, Int64, Tuple{}}}}})(flat::Vector{Float32})
    @ Optimisers ~/.julia/packages/Optimisers/UAVzc/src/destructure.jl:51

@mcabbott mcabbott linked a pull request Mar 26, 2022 that will close this issue
Draft
jondeuce added a commit to jondeuce/Optimisers.jl that referenced this issue Apr 21, 2022
@jondeuce
Offset wrapper to avoid confusing isleaf with `AbstractArray{Int}…
6a0ba9f 
…`s of offsets (also simplifying `_aux_children`); fix broken test for issue FluxML#62
@jondeuce jondeuce linked a pull request Apr 22, 2022 that will close this issue
Open
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
bug Something isn't working
Projects
None yet
Milestone
No milestone
Development

Successfully merging a pull request may close this issue.

Fix #62 jondeuce/Optimisers.jl
Attempt to fix #62 mcabbott/Optimisers.jl
3 participants
@rejuvyesh @ToucheSir @mcabbott