Closed pitx-perf closed 6 months ago
ChrisRackauckas
commented
6 months ago MOI can use symbolic information, and we should use it when we can. The best solution here is proably to try/catch then warn, and if it warns just have it omit the symbolic information. That will be slower but will make more code work (with an appropriate performance warning)
dmetivie
commented
6 months ago In a similar spirit,
Describe the example
OptimizationMOI from 0.1.16 to 0.2.0 introduces the use of Symbolics. Symbolics.Num is not compatible with Distributions.jl argument check.
Minimal Reproducible Example 👇
using Distributions
using Ipopt
using Optimization, OptimizationMOI
d = 1
T = 400
y = sin.(2π/T*1:T) + cos.(2π/T*1:T) .^ 2 .* rand(Gamma(), T)
f(θ) = Gamma(θ[1], θ[2])
f(t, θ) = f([cos(2π / T * t * θ[1])^2, cos(2π / T * t * θ[2])^2])
θσ10 = zeros(2d + 1)
θσ20 = zeros(2d + 1)
θ0 = hcat(θσ10, θσ20)
ℓ(θ, x) = -sum(logpdf(f(t / T, θ), x[t]) for t in eachindex(x)) # = -loglikelihood
OptFunc = OptimizationFunction(ℓ, Optimization.AutoForwardDiff())
prob = OptimizationProblem(OptFunc, vec(θ0), y)
sol = solve(prob, Ipopt.Optimizer())Error & Stacktrace ⚠️
ERROR: TypeError: non-boolean (Symbolics.Num) used in boolean context
Stacktrace:
[1] #271
@ Distributions C:\Users\metivier\.julia\packages\Distributions\UaWBm\src\univariate\continuous\gamma.jl:34 [inlined]
[2] check_args
@ Distributions C:\Users\metivier\.julia\packages\Distributions\UaWBm\src\utils.jl:89 [inlined]
[3] #Gamma#270
@ Distributions C:\Users\metivier\.julia\packages\Distributions\UaWBm\src\univariate\continuous\gamma.jl:34 [inlined]
[4] Gamma
@ Distributions C:\Users\metivier\.julia\packages\Distributions\UaWBm\src\univariate\continuous\gamma.jl:33 [inlined]
[5] f(θ::Vector{Symbolics.Num})
@ Main .\Untitled-1:10
[6] f(t::Float64, θ::Vector{Symbolics.Num})
@ Main .\Untitled-1:11
[7] (::var"#5#6"{Vector{Symbolics.Num}, Vector{Symbolics.Num}})(t::Int64)
@ Main .\none:0
[8] (::Base.MappingRF)(acc::Any, x::Any)
@ Base .\reduce.jl:100 [inlined]
[9] _foldl_impl(op::Base.MappingRF{var"#5#6"{…}, Base.BottomRF{…}}, init::Base._InitialValue, itr::Base.OneTo{Int64})
@ Base .\reduce.jl:58
[10] foldl_impl
@ .\reduce.jl:48 [inlined]
[11] mapfoldl_impl
@ .\reduce.jl:44 [inlined]
[12] mapfoldl
@ .\reduce.jl:175 [inlined]
[13] mapreduce
@ .\reduce.jl:307 [inlined]
[14] sum
@ .\reduce.jl:535 [inlined]
[15] sum
@ .\reduce.jl:564 [inlined]
[16] ℓ(θ::Vector{Symbolics.Num}, x::Vector{Symbolics.Num})
@ Main .\Untitled-1:18
[17] OptimizationFunction
@ C:\Users\metivier\.julia\packages\SciMLBase\dpafx\src\scimlfunctions.jl:3811 [inlined]
[18] modelingtoolkitize(prob::OptimizationProblem{…}; kwargs::@Kwargs{})
@ ModelingToolkit C:\Users\metivier\.julia\packages\ModelingToolkit\Gpzyo\src\systems\optimization\modelingtoolkitize.jl:19
[19] modelingtoolkitize
@ C:\Users\metivier\.julia\packages\ModelingToolkit\Gpzyo\src\systems\optimization\modelingtoolkitize.jl:6 [inlined]
[20] OptimizationMOI.MOIOptimizationNLPCache(prob::OptimizationProblem{…}, opt::Ipopt.Optimizer; callback::Nothing, kwargs::@Kwargs{…})
@ OptimizationMOI C:\Users\metivier\.julia\packages\OptimizationMOI\HC25i\src\nlp.jl:143
[21] MOIOptimizationNLPCache
@ C:\Users\metivier\.julia\packages\OptimizationMOI\HC25i\src\nlp.jl:108 [inlined]
[22] #__init#38
@ C:\Users\metivier\.julia\packages\OptimizationMOI\HC25i\src\OptimizationMOI.jl:278 [inlined]
[23] __init
@ C:\Users\metivier\.julia\packages\OptimizationMOI\HC25i\src\OptimizationMOI.jl:270 [inlined]
[24] #init#598
@ C:\Users\metivier\.julia\packages\SciMLBase\dpafx\src\solve.jl:165 [inlined]
[25] init
@ C:\Users\metivier\.julia\packages\SciMLBase\dpafx\src\solve.jl:163 [inlined]
[26] solve(::OptimizationProblem{…}, ::Ipopt.Optimizer; kwargs::@Kwargs{})
@ SciMLBase C:\Users\metivier\.julia\packages\SciMLBase\dpafx\src\solve.jl:96
[27] solve(::OptimizationProblem{…}, ::Ipopt.Optimizer)
@ SciMLBase C:\Users\metivier\.julia\packages\SciMLBase\dpafx\src\solve.jl:93
[28] top-level scope
@ Untitled-1:22
Some type information was truncated. Use `show(err)` to see complete types.Not Working Environment (please complete the following information):
[7f7a1694] Optimization v3.21.1
⌃ [fd9f6733] OptimizationMOI v0.3.1Working Environment (please complete the following information):
[7f7a1694] Optimization v3.21.1
⌃ [fd9f6733] OptimizationMOI v0.1.16Actually this MWE run with these versions but the solver fail (not error), maybe the MWE is ill posed. In my full problem, it worked.
Vaibhavdixit02
commented
6 months ago It should be fixed in the latest release which I created yesterday, try it out and let me know if it doesn't work!
ChrisRackauckas
commented
6 months ago I'm not sure about having it on by default without a time cap. There are many cases that would just stall which would otherwise work.
Vaibhavdixit02
commented
6 months ago It's off by default now
dmetivie
commented
6 months ago Seems like it is working! Thanks.
pitx-perf
commented
6 months ago Fixed for me as well. Thanks!
Describe the example
OptimizationMOI from 0.1.16 to 0.2.0 introduces the use of Symbolics. Symbolics.Num cannot be assigned to a MArray from StaticArrays because they are non-isbitstype.
Minimal Reproducible Example 👇
Error & Stacktrace ⚠️
Not Working Environment (please complete the following information):
using Pkg; Pkg.status()Working Environment (please complete the following information):
using Pkg; Pkg.status()