varmap_to_vars fails when used inside of loss function for optimization.

[mzhen77]

If I want to guarantee the order of parameters during the optimization I need to use varmap_to_vars, e.g:

loss function:

function loss(p::Vector{Float64})

    p1 = [param_vars[1] => p[1], param_vars[2] => p[2], param_vars[3] => p[3]]
    p2 = ModelingToolkit.varmap_to_vars(p1, param_vars)

    pred_sol = solve(prob, Tsit5(), p=p2, saveat=t_save)
    pred = Array(pred_sol')
    mse = mean((pred - obs_vals) .^ 2)
    return mse
end

where param_vars are the parameters of the MTK model.

The error I get:

ERROR: Compiling Tuple{Type{Dict}, Vector{Pair{Sym{Real, Base.ImmutableDict{DataType, Any}}, Float64}}}: try/catch is not supported. Refer to the Zygote documentation for fixes.

The full code description is here

[ChrisRackauckas]

If you do idxs = ModelingToolkit.varmap_to_vars(1:3, param_vars), then idxs should be the index map for how to permute p. So then inside of your loss function, you can use `

function loss(p::Vector{Float64})
    pred_sol = solve(prob, Tsit5(), p=p[idxs], saveat=t_save)
    pred = Array(pred_sol')
    mse = mean((pred - obs_vals) .^ 2)
    return mse
end

with idxs being constant. Or just use indexof to do this.

If this is a good workaround, then we should document this in the FAQ.

[mzhen77]

@ChrisRackauckas Thank you for your response.

But I can't understand it...

If I run the command as you adviced: idxs = ModelingToolkit.varmap_to_vars(1:3, param_vars) I get the error:
ERROR: MethodError: no method matching create_array(::Type{UnitRange{Int64}}, ::Type{Float64}, ::Val{1}, ::Val{3}, ::Float64, ::Float64, ::Float64)

If you mean something like that:

idxs = ModelingToolkit.varmap_to_vars([1,2,3], param_vars)
then the result will be [1,2,3] and doesn't depend on param_vars, so how can it help?

[ChrisRackauckas]

I just meant generate an index map.

using ModelingToolkit
p = @parameters x y z
idxs = ModelingToolkit.varmap_to_vars([p[1] => 1, p[2] => 2, p[3] => 3], p)
p[Int.(idxs)]

[mzhen77]

Thank you, @ChrisRackauckas.
It works. It is a good idea to document this in the FAQ.

[lamorton]

Edit: I'm dumb, I was just feeding it the actual defaults instead of the indexes.