How to train a biobjective problem with SDDP.jl?

z_yt_Yt · December 8, 2024, 2:10pm

Hello,

I am trying to run this SDDP.jl/papers/biobjective_sddp/BiObjectiveSDDP.jl at master · odow/SDDP.jl · GitHub and also other files in this folder SDDP.jl/papers/biobjective_sddp at master · odow/SDDP.jl · GitHub.

But I always get this warning, ERROR: UndefVarError: SingleVariable not defined in MathOptInterface
Suggestion: check for spelling errors or missing imports.

Anyone know how to solve this??
Thank you!!!

z_yt_Yt · December 8, 2024, 4:28pm

@odow Hi Oscar, could you please help me out here? Thank you!

odow · December 8, 2024, 8:10pm

This code was written for an older version of MathOptInterface.

I should have added version bounds to SDDP.jl/papers/biobjective_sddp/Project.toml at master · odow/SDDP.jl · GitHub, but I guess I didn’t

Regardless, I don’t recommend that you run the code. What are you trying to do and why? The biobjective algorithm works in theory, but not very well in practice.

z_yt_Yt · December 8, 2024, 9:31pm

Thank you for the reply!

Oh I saw this file but I didn’t realize they are relevant. I will try again.

I am writing the master’s thesis, and the topic is to incorporating carbon emissions in multi-stage lot sizing model with stochastic demand. It can be modeled in two ways: 1. minimize cost and carbon emissions as a constraint; 2. minimize cost and carbon emissions same time as a bi-objective model. Both are linear integer problem.
Thank you for asking! And I would appreciate any insights here on my topic

odow · December 8, 2024, 9:43pm

The most useful part of the paper was:

The idea of a bi-objective policy, and that you could use SDDP to build one
The heuristic solution method from the last section.

The least useful part of the paper was the exact solution algorithm, since it relied on the internals of the solver, and was highly susceptible to numerical issues.

I’ve implemented the heuristic as SDDP.train_biobjective, but it isn’t well documented yet.

Take a look at this example: Biobjective hydro-thermal · SDDP.jl

After training, you can use: SDDP.set_trade_off_weight(model, weight) to fix a particular weight and then simulate like you would with a single-objective problem.

Alternatively, you could also just solve a set of independent SDDP models for different scalarization weights.

z_yt_Yt · December 9, 2024, 10:28pm

Thank you for the valuable hints! I will think about them! Really appreciate that!

z_yt_Yt · January 12, 2025, 9:51pm

Hello Oscar @odow, I encountered another problem running the code. Here is part of the error message:

WARNING: replacing module BiObjectiveSDDP.
WARNING: redefinition of constant thermal_ub. This may fail, cause incorrect answers, or produce other errors.
WARNING: redefinition of constant thermal_lb. This may fail, cause incorrect answers, or produce other errors.
WARNING: redefinition of constant thermal_obj. This may fail, cause incorrect answers, or produce other errors.

It seems that there is something wrong about using the data in Brazilian_data.jl and I don’t know how to get it right.
Could you help me with this please? Thank you very much in advance!

odow · January 13, 2025, 12:59am

This warning happens when you include("/path/to/file.jl") a file multiple times. You shouldn’t do that.

z_yt_Yt · January 13, 2025, 11:15am

Thank you for the quick reply!

Strange, I have not changed anything in the .zip file I downloaded, but I am still getting this error.

I searched from “include” in the files in biobjective_SDDP, no dulplicates as expected. I tried “if !@isdefined(BiObjectiveSDDP)
include(joinpath(@DIR, “BiObjectiveSDDP.jl”))
end” as well, same error. I wonder what could be wrong here…

Besides, I should run “brazilian_example.jl” right? And it is supposed to generate some content like “simple_example.jl” does?

I really want to get this running, so far still trying haha

odow · January 13, 2025, 8:28pm

I assume that you are running this file multiple times:

github.com

odow/SDDP.jl/blob/master/papers/biobjective_sddp/brazilian_example.jl

#  Copyright (c) 2017-24, Oscar Dowson and SDDP.jl contributors.
#  This Source Code Form is subject to the terms of the Mozilla Public
#  License, v. 2.0. If a copy of the MPL was not distributed with this
#  file, You can obtain one at http://mozilla.org/MPL/2.0/.

include(joinpath(@__DIR__, "BiObjectiveSDDP.jl"))
using .BiObjectiveSDDP

using SDDP
import Gurobi
import Statistics

const OBJ_1_SCALING = 0.01
const OBJ_2_SCALING = 0.1
include("brazilian_data.jl")

function create_model(weight = nothing; stages = 60)
    env = Gurobi.Env()
    model = SDDP.LinearPolicyGraph(
        stages = stages,

This file has been truncated. show original

It has two includes:

github.com

odow/SDDP.jl/blob/aec55e2279c4abbd2a15ab1a430f6bb5a526976e/papers/biobjective_sddp/brazilian_example.jl#L6


      
          #  Copyright (c) 2017-24, Oscar Dowson and SDDP.jl contributors.
          #  This Source Code Form is subject to the terms of the Mozilla Public
          #  License, v. 2.0. If a copy of the MPL was not distributed with this
          #  file, You can obtain one at http://mozilla.org/MPL/2.0/.
          
          include(joinpath(@__DIR__, "BiObjectiveSDDP.jl"))
          using .BiObjectiveSDDP
          
          using SDDP
          import Gurobi
          import Statistics
          
          const OBJ_1_SCALING = 0.01
          const OBJ_2_SCALING = 0.1
          include("brazilian_data.jl")

github.com

odow/SDDP.jl/blob/aec55e2279c4abbd2a15ab1a430f6bb5a526976e/papers/biobjective_sddp/brazilian_example.jl#L15


      
          
          include(joinpath(@__DIR__, "BiObjectiveSDDP.jl"))
          using .BiObjectiveSDDP
          
          using SDDP
          import Gurobi
          import Statistics
          
          const OBJ_1_SCALING = 0.01
          const OBJ_2_SCALING = 0.1
          include("brazilian_data.jl")
          
          function create_model(weight = nothing; stages = 60)
              env = Gurobi.Env()
              model = SDDP.LinearPolicyGraph(
                  stages = stages,
                  lower_bound = 0.0,
                  optimizer = () -> Gurobi.Optimizer(env),
              ) do sp, t
                  set_silent(sp)
                  month = t % 12 == 0 ? 12 : t % 12  # Year to month conversion.

You must include these files only once.

including the file does not produce output by default. It was intended to be run from the command line:

github.com

odow/SDDP.jl/blob/aec55e2279c4abbd2a15ab1a430f6bb5a526976e/papers/biobjective_sddp/brazilian_example.jl#L274-L314


      
          function help()
              println("""julia brazilian_example.jl --experiment={1,2,3} -n 9 -atol 100
          
              ## Arguments
          
               * --experiment :: choose which experiment to Run
          
              ### Experiment 2
          
               * -n    :: Choose how many weights to run
               * -atol :: The tolerance used by the `BoundStalling` stopping rule
          
              ## Examples
          
              ```
              nohup ~/julia1.6 --project=. brazilian_example.jl --experiment=1 &
              nohup ~/julia1.6 --project=. brazilian_example.jl --experiment=2 -n 9 -atol 1e2 &
              nohup ~/julia1.6 --project=. brazilian_example.jl --experiment=3 &
              ```
              """)

This file has been truncated. show original

But I will caution you again: I don’t think there is value in attempting to run this code.

z_yt_Yt · January 14, 2025, 11:49pm

Thank you so much for the detailed answer!

I understood, I will definitely try to implement my model using the heuristic.

Really appreciate your help!

Topic		Replies	Views
Method Error with no method matching zero(::Type{SDDP.State{VariableRef}}) Optimization (Mathematical)	2	29	September 6, 2024
Nonlinear constraints in SDDP Optimization (Mathematical) jump , sddp	3	243	October 31, 2023
Announcing new Packages and Solvers for Multi-objective Mixed Integer Programming ( with JuMP extension ) Optimization (Mathematical)	4	2991	September 11, 2017
SDDP:MethodError: no method matching *(::Float64, ::SDDP.State{VariableRef}) Optimization (Mathematical)	4	481	May 23, 2021
Bug when using DSDP versus SCS optimiser with JuMP Optimization (Mathematical) question , jump , optimization	2	460	March 28, 2020

How to train a biobjective problem with SDDP.jl?

Related topics