Results from our experiments

Result data

The full results from our computational experiments are available in the folder Computational_Results. For a description on how to interpret the output files, see section logging.

Reproducing the results

To reproduce the results from our paper, it is simply required to run the files run_model.jl, run_model_linearized.jl and run_model_markov.jl on the current main branch.

Note that for loglinearSDDP, cut selection is not supported yet, so one run of run_model.jl is sufficient. For the other two variants, we executed two batches of runs, one with and one without cut selection. The default case is that SDDP.jl is run with cut selection in both run_model_linearized.jl and run_model_markov.jl.

Unfortunately, the train function in SDDP.jl does not have an argument to control the cut selection scheme. Therefore, if run_model_linearized.jl and run_model_markov.jl should be run without cut selection, it is required to develop the SDDP.jl package and to manually switch the cut_selection parameter from True to False in the file bellman_functions.jl for function add_cut.jl.

Keep in mind that we used Gurobi 11 and older versions of JuMP.jl, SDDP.jl and several other packages when performing our experiments, so the results that you obtain may slightly deviate from what we report. The same can be true if the code is executed on a different hardware than ours.

This page was generated using Literate.jl.