EnsembleProblem support AbstractVector

[Beforerr]

Is your feature request related to a problem? Please describe.

Rightnow EnsembleProblem(prob::AbstractSciMLProblem, u0s::Vector{Vector{T}}; kwargs...) only support Vector{Vector{T}} as u0s, but in some cases, we may want to use Vector{StaticVector} as u0s.

Describe the solution you’d like

So I think it would be better to support AbstractVector as u0s.

[ChrisRackauckas]

I don't understand. EnsembleProblem allows for static array problems via prob_func. Can you share an example/MWE?

[Beforerr]

Yes, using prob_func would works. But the EnsembleProblem(prob, u0s) interface would not work.

using OrdinaryDiffEq
using StaticArrays

function lorenz(u, p, t)
    du1 = 10.0 * (u[2] - u[1])
    du2 = u[1] * (28.0 - u[3]) - u[2]
    du3 = u[1] * u[2] - (8 / 3) * u[3]
    return [du1, du2, du3]
end

# working
u0s = [randn(3) for i in 1:100]
prob = ODEProblem(lorenz, u0s[1], (0.0, 1.0))
ensemble_prob = EnsembleProblem(prob, u0s)
sim = solve(ensemble_prob, Tsit5(), EnsembleThreads(), trajectories=length(u0s))

# not working
function slorenz(u, p, t)
    du1 = 10.0 * (u[2] - u[1])
    du2 = u[1] * (28.0 - u[3]) - u[2]
    du3 = u[1] * u[2] - (8 / 3) * u[3]
    return SA[du1, du2, du3]
end

su0s = [SVector{3}(u0) for u0 in u0s]
prob = ODEProblem(slorenz, u0s[1], (0.0, 1.0))
ensemble_prob = EnsembleProblem(prob, su0s)

[ChrisRackauckas]

What interface? share the docs link?

[Beforerr]

You may check the code here.

https://github.com/SciML/SciMLBase.jl/blob/babd23945f2fa2ac8b0ed6c77336ec6a7ecaad66/src/ensemble/ensemble_problems.jl#L48

[ChrisRackauckas]

The u0s thing isn't documented and is deprecated before it ever got documented though?

[Beforerr]

Thanks for clarification!

[ChrisRackauckas]

Okay so this wasn't pulled from any documentation? If so, then this is at least a completion to the issue since there hasn't been an intent to finish it. If it's just from reading the code, note that this interface is not planned to be finished, it's planned to be removed since there's some issues with it and it does not save much over just using the prob_func interface.

[ChrisRackauckas]

But if this is documented anywhere, please please let me know, that's not intended.

[Beforerr]

Thanks. I do not think it is documented anywhere.