Skip to content

DecisionProgramming.jl is a Julia package for solving multi-stage decision problems under uncertainty, modeled using influence diagrams. Internally, it relies on mathematical optimization. Decision models can be embedded within other optimization models.

License

Notifications You must be signed in to change notification settings

toubinaattori/DecisionProgramming.jl

 
 

Repository files navigation

DecisionProgramming.jl

Docs Image Runtests DOI

Description

DecisionProgramming.jl is a Julia package for solving multi-stage decision problems under uncertainty, modeled using influence diagrams. Internally, it relies on mathematical optimization. Decision models can be embedded within other optimization models. We designed the package as JuMP extension. We have also developed a Python interface, which is available here.

Citing

The Decision Programming framework is described in this publication. If you found the framework useful in your work, we kindly ask you to cite the following publication (pdf):

@article{Salo_et_al-2022,
    title = {Decision programming for mixed-integer multi-stage optimization under uncertainty},
    journal = {European Journal of Operational Research},
    volume = {299},
    number = {2},
    pages = {550-565},
    year = {2022},
    issn = {0377-2217},
    doi = {https://doi.org/10.1016/j.ejor.2021.12.013},
    url = {https://www.sciencedirect.com/science/article/pii/S0377221721010201},
    author = {Ahti Salo and Juho Andelmin and Fabricio Oliveira},
    keywords = {Decision analysis, Influence diagrams, Decision trees, Contingent portfolio programming, Stochastic programming}
}

If you use the rooted junction tree models, we kindly ask you to cite the following publication (pdf):

@article{Parmentier_et_al-2020,
    title = {Integer programming on the junction tree polytope for influence diagrams},
    journal = {INFORMS Journal on Optimization},
    volume = {2},
    number = {3},
    pages = {209--228},
    year = {2020},
    issn = {0377-2217},
    doi = {https://doi.org/10.1287/ijoo.2019.0036},
    url = {https://pubsonline.informs.org/doi/epdf/10.1287/ijoo.2019.0036},
    author = {Parmentier, Axel and Cohen, Victor and Lecl{\`e}re, Vincent and Obozinski, Guillaume and Salmon, Joseph},
    keywords = {Influence diagrams, Partially observed Markov decision processes, probabilistic graphical models, Linear programming}
}

Syntax

We can create an influence diagram as follows:

using DecisionProgramming

diagram = InfluenceDiagram()

add_node!(diagram, DecisionNode("A", [], ["a", "b"]))
add_node!(diagram, ChanceNode("B", ["A"], ["x", "y"]))
add_node!(diagram, ChanceNode("C", ["A"], ["v", "w"]))
add_node!(diagram, DecisionNode("D", ["B", "C"], ["k", "l"]))
add_node!(diagram, ValueNode("V", ["D"]))

generate_arcs!(diagram)

add_probabilities!(diagram, "B", [0.4 0.6; 0.6 0.4])
add_probabilities!(diagram, "C", [0.7 0.3; 0.3 0.7])
add_utilities!(diagram, "V", [1.5, 1.7])

Using the influence diagram, we generate the model as follows:

using JuMP
model, z, variables = generate_model(diagram, model_type="RJT")

We can optimize the model using MILP solver.

using HiGHS
optimizer = optimizer_with_attributes(
    () -> HiGHS.Optimizer()
)
set_optimizer(model, optimizer)
optimize!(model)

Finally, we extract the decision strategy from the decision variables.

Z = DecisionStrategy(z)

See the documentation for more detailed examples.

Installation

DecisionProgramming.jl is registered. You can add it using the command:

pkg> add DecisionProgramming

To run examples and develop and solve decision models, you have to install JuMP and a solver capable of solving mixed-integer linear programs (MILP). JuMP documentation contains a list of available solvers.

pkg> add JuMP

We recommend using the HiGHS solver, which is an efficient open-source solver.

pkg> add HiGHS

Now you are ready to use decision programming.

Development

Using the package manager, add DecisionProgramming.jl package for development using the command:

pkg> develop DecisionProgramming

If you have already cloned DecisionProgramming from GitHub, you can use the command:

pkg> develop .

Inside DecisionProgramming directory, run tests using the commands:

pkg> activate .
(DecisionProgramming) pkg> test

You can find more instruction on how to install packages for development at Julia's Pkg documentation.

About

DecisionProgramming.jl is a Julia package for solving multi-stage decision problems under uncertainty, modeled using influence diagrams. Internally, it relies on mathematical optimization. Decision models can be embedded within other optimization models.

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Julia 100.0%