print a state variable (var.in)

[Esnilg]

hello oscar

One question, is there any way to print a state variable (var.in) during the SDDP train?

[odow]

You could imagine doing something like:

SDDP.LinearPolicyGraph() do sp, node
    @variable(sp, var >= 0, SDDP.State, initial_value = 0)
    SDDP.parameterize(sp, 1:3) do w
        v = fix_value(var.in)
        println("In $(node) with w=$(w), got var=$(v)")
        @stageobjective(sp, w + var.out)
    end
end

Why do you want to do this though? Do you just want to print the forward pass?

There is this wildly undocumented pre_optimize_hook that you can set https://github.com/odow/SDDP.jl/commit/88af622304a44a6481485dfaf2f6ca15255143f4 which has access to the incoming states

Something like this

model = SDDP.LinearPolicyGraph() do sp, node
    @variable(sp, var >= 0, SDDP.State, initial_value = 0)
    SDDP.parameterize(sp, 1:3) do w
        v = fix_value(var.in)
        println("In $(node) with w=$(w), got var=$(v)")
        @stageobjective(sp, w + var.out)
    end
end

for (k, node) in model.nodes
    SDDP.pre_optimize_hook(node) do model, node, state, noise, scenario_path, require_duals
         @show state
    end
end

[odow]

Closing because this seems resolved. There isn't really a good reason to print the state variable during training, so it's not something I want to support long-term.