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.

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

No method matching Duplicated(::Base.MappingRF{...}, ::Base.RefValue{Base.MappingRF{…}}) #1812

Open
mhauru opened this issue Sep 12, 2024 · 13 comments

Comments

@mhauru
Copy link
Contributor

mhauru commented Sep 12, 2024

MWE:

using Random: randn
using Enzyme: Enzyme
using Turing: Turing

Enzyme.API.runtimeActivity!(true)

Turing.@model function MvDirichletWithManualAccumulation(w, doc)
    β ~ Turing.filldist(Turing.Dirichlet([1.0, 1.0]), 2)
    log_product = log.(β)
    Turing.@addlogprob! sum(log_product[CartesianIndex.(w, doc)])
end

model = MvDirichletWithManualAccumulation([1, 1, 1, 1], [1, 1, 2, 2])
ctx = Turing.DefaultContext()
vi = Turing.VarInfo(model)
vi_linked = Turing.link(vi, model)
ldp = Turing.LogDensityFunction(vi_linked, model, ctx)
f = Base.Fix1(Turing.LogDensityProblems.logdensity, ldp)
d = Turing.LogDensityProblems.dimension(ldp)
x = randn(d)

Enzyme.gradient(Enzyme.Reverse, Enzyme.Const(f), x)

Output:

ERROR: LoadError: MethodError: no method matching EnzymeCore.Duplicated(::Base.MappingRF{Base.Fix1{…}, Base.BottomRF{…}}, ::Base.RefValue{Base.MappingRF{…}})

Closest candidates are:
  EnzymeCore.Duplicated(::T1, ::T1) where T1
   @ EnzymeCore ~/.julia/packages/EnzymeCore/jSKtE/src/EnzymeCore.jl:66
  EnzymeCore.Duplicated(::T1, ::T1, ::Bool) where T1
   @ EnzymeCore ~/.julia/packages/EnzymeCore/jSKtE/src/EnzymeCore.jl:66

Stacktrace:
  [1] runtime_generic_augfwd(activity::Type{…}, width::Val{…}, ModifiedBetween::Val{…}, RT::Val{…}, f::Base.MappingRF{…}, df::Base.RefValue{…}, primal_1::Base._InitialValue, shadow_1_1::Nothing, primal_2::SubArray{…}, shadow_2_1::SubArray{…})
    @ Enzyme.Compiler ~/.julia/packages/Enzyme/TiboG/src/rules/jitrules.jl:338
  [2] _foldl_impl
    @ ./reduce.jl:58
  [3] foldl_impl
    @ ./reduce.jl:48 [inlined]
  [4] foldl_impl
    @ ./reduce.jl:0 [inlined]
  [5] augmented_julia_foldl_impl_8766_inner_1wrap
    @ ./reduce.jl:0
  [6] macro expansion
    @ ~/.julia/packages/Enzyme/TiboG/src/compiler.jl:7187 [inlined]
  [7] enzyme_call
    @ ~/.julia/packages/Enzyme/TiboG/src/compiler.jl:6794 [inlined]
  [8] AugmentedForwardThunk
    @ ~/.julia/packages/Enzyme/TiboG/src/compiler.jl:6682 [inlined]
  [9] runtime_generic_augfwd(activity::Type{…}, width::Val{…}, ModifiedBetween::Val{…}, RT::Val{…}, f::typeof(Base.foldl_impl), df::Nothing, primal_1::Base.MappingRF{…}, shadow_1_1::Base.RefValue{…}, primal_2::Base._InitialValue, shadow_2_1::Nothing, primal_3::Distributions.EachVariate{…}, shadow_3_1::Distributions.EachVariate{…})
    @ Enzyme.Compiler ~/.julia/packages/Enzyme/TiboG/src/rules/jitrules.jl:338
 [10] mapfoldl_impl
    @ ./reduce.jl:44 [inlined]
 [11] mapfoldl
    @ ./reduce.jl:175 [inlined]
 [12] _mapreduce
    @ ./reduce.jl:453 [inlined]
 [13] _mapreduce_dim
    @ ./reducedim.jl:365 [inlined]
 [14] mapreduce
    @ ./reducedim.jl:357 [inlined]
 [15] #_sum#831
    @ ./reducedim.jl:1015 [inlined]
 [16] augmented_julia___sum_831_8532_inner_1wrap
    @ ./reducedim.jl:0
 [17] macro expansion
    @ ~/.julia/packages/Enzyme/TiboG/src/compiler.jl:7187 [inlined]
 [18] enzyme_call
    @ ~/.julia/packages/Enzyme/TiboG/src/compiler.jl:6794 [inlined]
 [19] AugmentedForwardThunk
    @ ~/.julia/packages/Enzyme/TiboG/src/compiler.jl:6682 [inlined]
 [20] runtime_generic_augfwd(activity::Type{…}, width::Val{…}, ModifiedBetween::Val{…}, RT::Val{…}, f::Base.var"##_sum#831", df::Nothing, primal_1::@Kwargs{}, shadow_1_1::Nothing, primal_2::typeof(Base._sum), shadow_2_1::Nothing, primal_3::Base.Fix1{…}, shadow_3_1::Base.RefValue{…}, primal_4::Distributions.EachVariate{…}, shadow_4_1::Distributions.EachVariate{…}, primal_5::Colon, shadow_5_1::Nothing)
    @ Enzyme.Compiler ~/.julia/packages/Enzyme/TiboG/src/rules/jitrules.jl:338
 [21] _sum
    @ ./reducedim.jl:1015 [inlined]
 [22] sum
    @ ./reducedim.jl:1011 [inlined]
 [23] loglikelihood
    @ ~/.julia/packages/Distributions/nmUhn/src/common.jl:458 [inlined]
 [24] _logpdf
    @ ~/.julia/packages/DistributionsAD/1Mxge/src/filldist.jl:97 [inlined]
 [25] logpdf
    @ ~/.julia/packages/Distributions/nmUhn/src/common.jl:263 [inlined]
 [26] invlink_with_logpdf
    @ ~/.julia/packages/DynamicPPL/DvdZw/src/abstract_varinfo.jl:856
 [27] invlink_with_logpdf
    @ ~/.julia/packages/DynamicPPL/DvdZw/src/abstract_varinfo.jl:850 [inlined]
 [28] assume
    @ ~/.julia/packages/DynamicPPL/DvdZw/src/context_implementations.jl:227 [inlined]
 [29] tilde_assume
    @ ~/.julia/packages/DynamicPPL/DvdZw/src/context_implementations.jl:60 [inlined]
 [30] tilde_assume
    @ ~/.julia/packages/DynamicPPL/DvdZw/src/context_implementations.jl:57 [inlined]
 [31] tilde_assume!!
    @ ~/.julia/packages/DynamicPPL/DvdZw/src/context_implementations.jl:144
 [32] MvDirichletWithManualAccumulation
    @ ~/projects/Enzyme-mwes/MappingRF_BottomRF/mwe.jl:11
 [33] _evaluate!!
    @ ~/.julia/packages/DynamicPPL/DvdZw/src/model.jl:973 [inlined]
 [34] evaluate_threadunsafe!!
    @ ~/.julia/packages/DynamicPPL/DvdZw/src/model.jl:946 [inlined]
 [35] evaluate!!
    @ ~/.julia/packages/DynamicPPL/DvdZw/src/model.jl:894 [inlined]
 [36] logdensity
    @ ~/.julia/packages/DynamicPPL/DvdZw/src/logdensityfunction.jl:140
 [37] Fix1
    @ ./operators.jl:1118 [inlined]
 [38] Fix1
    @ ./operators.jl:0 [inlined]
 [39] diffejulia_Fix1_10564_inner_2wrap
    @ ./operators.jl:0
 [40] macro expansion
    @ ~/.julia/packages/Enzyme/TiboG/src/compiler.jl:7187 [inlined]
 [41] enzyme_call
    @ ~/.julia/packages/Enzyme/TiboG/src/compiler.jl:6794 [inlined]
 [42] CombinedAdjointThunk
    @ ~/.julia/packages/Enzyme/TiboG/src/compiler.jl:6671 [inlined]
 [43] autodiff
    @ ~/.julia/packages/Enzyme/TiboG/src/Enzyme.jl:320 [inlined]
 [44] gradient(rm::EnzymeCore.ReverseMode{…}, f::EnzymeCore.Const{…}, x::Vector{…})
    @ Enzyme ~/.julia/packages/Enzyme/TiboG/src/Enzyme.jl:1049
 [45] top-level scope
    @ ~/projects/Enzyme-mwes/MappingRF_BottomRF/mwe.jl:25
 [46] include(fname::String)
    @ Base.MainInclude ./client.jl:489
 [47] top-level scope
    @ REPL[3]:1
in expression starting at /Users/mhauru/projects/Enzyme-mwes/MappingRF_BottomRF/mwe.jl:1
Some type information was truncated. Use `show(err)` to see complete types.
@wsmoses
Copy link
Member

wsmoses commented Sep 12, 2024 via email

@mhauru
Copy link
Contributor Author

mhauru commented Sep 12, 2024

Somewhat simplified:

using Random: randn
using Enzyme: Enzyme
using DynamicPPL: DynamicPPL
using Distributions: Dirichlet
using DistributionsAD: filldist

Enzyme.API.runtimeActivity!(true)

w = [1, 1, 1, 1]
doc = [1, 1, 2, 2]
vn = DynamicPPL.VarName{:β}()
vi_template = DynamicPPL.VarInfo()
d = filldist(Dirichlet([1.0, 1.0]), 2)
β = rand(2,2)
vi = DynamicPPL.push!!(vi_template, vn, β, d, Set{DynamicPPL.Selector}())

function f(x)
    vi = DynamicPPL.unflatten(vi_template, x)
    ctx = DynamicPPL.DefaultContext()
    β, vi = DynamicPPL.tilde_assume!!(
        ctx, filldist(Dirichlet([1.0, 1.0]), 2), vn, vi
    )
    log_product = log.(β)
    vi = DynamicPPL.acclogp!!(ctx, vi, sum(log_product[CartesianIndex.(w, doc)]))
    return DynamicPPL.getlogp(vi)
end

x = randn(4)

Enzyme.gradient(Enzyme.Reverse, Enzyme.Const(f), x)

@wsmoses
Copy link
Member

wsmoses commented Sep 12, 2024

Hm sorry, but could you give it a bit more minimization still?

@wsmoses
Copy link
Member

wsmoses commented Sep 12, 2024

actually wait, nevermind

@mhauru
Copy link
Contributor Author

mhauru commented Sep 13, 2024

Can do more if it helps, the above is just what I got to yesterday before calling it a day.

@wsmoses
Copy link
Member

wsmoses commented Sep 24, 2024

With the DynamicPPL release out, closing this

@wsmoses wsmoses closed this as completed Sep 24, 2024
@mhauru
Copy link
Contributor Author

mhauru commented Sep 24, 2024

Sorry @wsmoses, I don't understand. This still fails with the same error on the latest DynamicPPL and Enzyme main. Reopen?

Above MWE with the new way of setting runtime activity:

using Random: randn
using Enzyme: Enzyme
using DynamicPPL: DynamicPPL
using Distributions: Dirichlet
using DistributionsAD: filldist

w = [1, 1, 1, 1]
doc = [1, 1, 2, 2]
vn = DynamicPPL.VarName{:β}()
vi_template = DynamicPPL.VarInfo()
d = filldist(Dirichlet([1.0, 1.0]), 2)
β = rand(2,2)
vi = DynamicPPL.push!!(vi_template, vn, β, d, Set{DynamicPPL.Selector}())

function f(x)
    vi = DynamicPPL.unflatten(vi_template, x)
    ctx = DynamicPPL.DefaultContext()
    β, vi = DynamicPPL.tilde_assume!!(
        ctx, filldist(Dirichlet([1.0, 1.0]), 2), vn, vi
    )
    log_product = log.(β)
    vi = DynamicPPL.acclogp!!(ctx, vi, sum(log_product[CartesianIndex.(w, doc)]))
    return DynamicPPL.getlogp(vi)
end

x = randn(4)

Enzyme.gradient(Enzyme.set_runtime_activity(Enzyme.Reverse), Enzyme.Const(f), x)

@wsmoses
Copy link
Member

wsmoses commented Sep 24, 2024

oh sorry misremembered this one and the other one

@wsmoses wsmoses reopened this Sep 24, 2024
@wsmoses
Copy link
Member

wsmoses commented Sep 24, 2024

and yeah okay sadly this needs further reduction as well

@mhauru
Copy link
Contributor Author

mhauru commented Oct 4, 2024

Reduced:

module MWE

using Enzyme: Enzyme
using Distributions: Distributions

Enzyme.API.runtimeActivity!(true)

function f(x)
    d = Distributions.Dirichlet([1.0, 1.0])
    v = eachrow(x)
    return sum(Base.Fix1(Distributions.logpdf, d), v)
end

x = reshape([0.8, 0.3, 0.2, 0.7], (2, 2))
@show f(x)
Enzyme.gradient(Enzyme.Reverse, f, x)

end

@wsmoses
Copy link
Member

wsmoses commented Oct 4, 2024 via email

@mhauru
Copy link
Contributor Author

mhauru commented Oct 4, 2024

Same error on Enzyme v0.13.7 (once you remove the runtimeActivity! setting).

I gotta go attend to other things, feel free to minimise further if you have time.

@wsmoses
Copy link
Member

wsmoses commented Nov 4, 2024

@penelopeysm if you have time, would you be able to reduce this one as well?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

No branches or pull requests

2 participants