Add a settable optimize_hook with access to the current scenario.

[odow]

No description provided.

[codecov]

Codecov Report

Merging #219 into master will decrease coverage by 0.02%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #219      +/-   ##
==========================================
- Coverage    93.3%   93.27%   -0.03%     
==========================================
  Files          11       12       +1     
  Lines        1075     1101      +26     
==========================================
+ Hits         1003     1027      +24     
- Misses         72       74       +2

Impacted Files	Coverage Δ
src/algorithm.jl	98.24% <100%> (-0.2%)	⬇️
src/user_interface.jl	97.26% <100%> (+0.04%)	⬆️
src/SDDP.jl	100% <0%> (ø)	⬆️
src/plugins/headers.jl	100% <0%> (ø)	⬆️
src/visualization/dashboard.jl	0% <0%> (ø)
src/visualization/spaghetti_plot.jl	82.6% <0%> (+19.27%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 33e7024...f9550cf. Read the comment docs.

[peraaslid]

Can I use this to add new restriction before/after each iteration? I'd like to add DC powerflow active losses as "cuts" iterative. Could you give an example of how to use the pre-/post-optimize hook?

[odow]

The current example is here:

SDDP.jl/test/algorithm.jl

Lines 150 to 176 in 9b5ebc9

	@testset "optimize_hook" begin
	model = SDDP.LinearPolicyGraph(
	stages = 2,
	optimizer = with_optimizer(GLPK.Optimizer),
	lower_bound = 0.0
	) do sp, t
	@variable(sp, x >= 0, SDDP.State, initial_value = 0)
	@stageobjective(sp, x.out)
	end
	pre_optimize_called = 0
	post_optimize_called = 0
	node = model[1]
	SDDP.pre_optimize_hook(node) do model, node, state, noise, scenario_path, require_duals
	pre_optimize_called = 1
	return pre_optimize_called
	end
	SDDP.post_optimize_hook(node) do ret
	post_optimize_called = ret + 2
	return
	end
	SDDP.solve_subproblem(
	model, node, Dict(:x => 0.0), nothing, Tuple{Int, Any}[(1, nothing)];
	require_duals = false
	)
	@test pre_optimize_called == 1
	@test post_optimize_called == 3
	end

For each node in the tree, you can add a hook. To loop through nodes, you can go

for (index, node) in model.nodes
    # add hook
end

pre_optimize hook is a function that takes the same arguments as solve_subproblem. If it returns a value, and post_optimize_hook is set, then the returned value will be passed as a the sole argument to post_optimize_hook.

function my_pre_optimize_hook(model, node, state, noise, scenario_path, require_duals)
    sp = node.subproblem
    active_loss_constraint = @constraint(sp, sp[:x] <= 1) 
    return active_loss_constraint
end

function my_post_optimize_hook(ret)
    # `ret` is the `active_loss_constraint` reference returned from `my_pre_optimize_hook`
    println("Active loss constraint = $(ret)")
    return
end

for (index, node) in model.nodes
    SDDP.pre_optimize_hook(my_pre_optimize_hook, node)
    SDDP.post_optimize_hook(my_post_optimize_hook, node)
end

Note that post_optimize_hook doesn't get called immediately after JuMP.optimize!. I need that for my own work, but I could be compelled to change it if you need.

SDDP.jl/src/algorithm.jl

Lines 293 to 295 in 9b5ebc9

	if node.post_optimize_hook !== nothing
	node.post_optimize_hook(pre_optimize_ret)
	end