Kirill Zubov

I have problem with RuntimeGeneratedFunctions. ```julia using RuntimeGeneratedFunctions, DiffEqFlux, ForwardDiff phi = FastChain(FastDense(1,12,Flux.tanh),FastDense(12,12,Flux.tanh),FastDense(12,1)) indvars = [:x] undv = [1] derivative_expr = parser_derivative(phi,indvars,undv) derivative_expr =:((θ_, x)->begin #= none:3 =# ForwardDiff.derivative((x->begin #=...

I still can't figure out how to use the @RuntimeGeneratedFunction, here one way or another it needs to call an anonymous function inside the function body. Which leads to an...

Derivative benchmarks. ```julia chain = FastChain(FastDense(2,16,Flux.σ),FastDense(16,16,Flux.σ),FastDense(16,1)) initθ = Float64.(DiffEqFlux.initial_params(chain)) eltypeθ = eltype(initθ) parameterless_type_θ = DiffEqBase.parameterless_type(initθ) phi = NeuralPDE.get_phi(chain,parameterless_type_θ) u_ = (cord, θ, phi)->phi(cord, θ) _epsilon = one(eltype(initθ)) / (2*cbrt(eps(eltype(initθ)))) ε...

Poisson equation benchmarks ```julia @parameters x y @variables u(..) Dxx = Differential(x)^2 Dyy = Differential(y)^2 eq = Dxx(u(x,y)) + Dyy(u(x,y)) ~ -sin(pi*x)*sin(pi*y) bcs = [u(0,y) ~ 0.0, u(1,y) ~ -sin(pi*1)*sin(pi*y),...

dependency runtime of the size of the neural network ```julia num = ... fastchain = FastChain(FastDense(2,num,Flux.σ),FastDense(num,num,Flux.σ),FastDense(num,1)) initθ = Float64.(DiffEqFlux.initial_params(fastchain)) grid_strategy = NeuralPDE.GridTraining(0.1) quasirandom_strategy = NeuralPDE.QuasiRandomTraining(50; bcs_points= 10, sampling_alg = LatinHypercubeSample())...

Poisson's equation prediction accuracy automatic / numerical differentiation ```julia num = 5 fastchain = FastChain(FastDense(2,num,Flux.σ),FastDense(num,num,Flux.σ),FastDense(num,1)) initθ = Float64.(DiffEqFlux.initial_params(fastchain)) grid_strategy = NeuralPDE.GridTraining(0.05) discretization = NeuralPDE.PhysicsInformedNN(fastchain, grid_strategy; init_params = initθ, AD =false)...

problem with opaque_closure ```julia julia> VERSION v"1.7.1" chain = FastChain(FastDense(1,16,Flux.σ),FastDense(16,16,Flux.σ),FastDense(16,1)) initθ = Float64.(DiffEqFlux.initial_params(chain)) phi_ = NeuralPDE.get_phi(chain,DiffEqBase.parameterless_type(initθ)) u_(x,θ_) = phi_(vcat(x),θ_)[1] u_(1, initθ) inner_expr =:(x->$u_(x,θ_)) inner_func = @RuntimeGeneratedFunction(inner_expr) expr = :((x,θ_)->ForwardDiff.derivative($inner_expr,x)) func...

@ChrisRackauckas do you have any ideas on how to deal with this?

Add Dual and Tags for Core.OpaqueClosure?

yes, it is the same error ```julia using RuntimeGeneratedFunctions RuntimeGeneratedFunctions.init(@__MODULE__) chain = FastChain(FastDense(1,16,Flux.σ),FastDense(16,16,Flux.σ),FastDense(16,1)) initθ = Float64.(DiffEqFlux.initial_params(chain)) phi_ = NeuralPDE.get_phi(chain,DiffEqBase.parameterless_type(initθ)) u_(x,θ_) = phi_(vcat(x),θ_)[1] u_(1, initθ) inner_expr =:(θ->(x-> $u_(x,θ))) #nested inner_func =...