Price interpolation and Markov states

[odow]

A user writes

In an example of your work "Stochastic dual dynamic programming with stagewise dependent objective uncertainty”, you use a value_function = priceprocess(USE_AR1) as a parameter of SDDPModel. I want to know if this we can modify the priceprocess function to allow noises from different discrete distributions. For example, think of a tree with different branches at stage 2 than at stage 3.

Here is an example

const DISTRIBUTIONS = [
    [1, 2, 3], # distributions in Markov state 1
    [3, 4, 5]  # distributions in Markov state 2
]
function buildvaluefunction(stage::Int, markov::Int)
    return DynamicPriceInterpolation(
        # ... other arguments omitted ...
        dynamics = (price, noise) -> price + noise,
        noises = DISTRIBUTIONS[markov]
    )
end
m = SDDPModel(
    stages = 3,
    # The transition matrix
    #    x - x
    #   / \ /
    # x    X
    #   \ / \
    #    x - x
    markov_transition = [
        [1.0]'
        [0.5 0.5],
        [0.5 0.5; 0.5 0.5]
    ],
    # pass a function that takes (stage, markov state) as arguments 
    #    and returns a value function.
    value_functon = buildvaluefunction
        ) do sp, t, i
    # you can also use the state t, and markov state i,
    #     inside the subproblem, e.g.
    @state(sp, x >= 2 * t + i, x0==0)
    @stageobjective(sp, p-> (t + i) * p * x
end

[lorenzoreus]

Dear Oscar, I'm trying to apply previous code, but I can't make it work. buildfunction is defined with arguments, but no argumets are used when it is used in the SDDPModel. I've tried doing some changes from this example with no success. Any help is appreciated. Nice package by the way.

[odow]

Lorenzo,

See #122.

Nice package by the way.

Thanks!

[lorenzoreus]

Thanks so much! It works