Understanding the logging output

For each run in our experiments, several outputs are logged in a dedicated log-file. It consists of the following elements:

• A section printing the main model run parameters and some general information
  • the date and the time of the run
  • the main algorithmic parameters stored in AlgoParams (stopping rules, regularization_regime, vector of cut_generation_regime including the dual_regime with all parameters and the state_approximation_regime, cut_aggregation_regime, cut_selection_regime, late binarization information (no longer part of the code))
  • the applied solvers and solver parameters
  • the seed for the simulation
  • an optional run description by the user
  • the main problem parameters stored in ProblemParams (number of stages, number of realizations per stage, seed for th forward pass)
• A section logging information from the SDDiP iterations. Each row contains
  • the iteration number
  • the upper bound estimator for the respective iteration (this is a very bad estimator and only relevant for deterministic problems)
  • the max upper bound estimator found so far (this is a very bad estimator and only relevant for deterministic problems)
  • the deterministic lower bound
  • the optimality gap (this is only relevant for deterministic problems)
  • the deterministic lower bound
  • the total time in seconds
  • the time in seconds for the respective iteration
  • some information on refinements of the regularization parameter or the state binarization that took place (if Regularization and BinaryApproximation are used at all)
  • the total number of variables in each subproblem
  • the number of binary variables in each subproblem
  • the total number of original constraints in each subproblem
  • the total number of cuts in the subproblems (aggregated over all stages)
  • the active number of cuts in the subproblems (aggregated over all stages; relevant if CutSelection is used)
  • some information on the total number of iterations required to solve all the Lagrangian dual problems (if any) in the respective iteration
• A table summarizing timing and memory allocation information for different steps of the algorithm
• A summary of the final statuses of solving the Lagrangian dual problems (if any)
  • opt: the problem was solved to optimality
  • iter: the solution stopped at the iteration limit
  • conv: the solution method converged to a suboptimal solution
  • unbounded: the problem was detected as unbounded (relevant for linear normalization); a SB cut is generated instead
  • bound_issues or feas_issues: the solution method showed some numerical issues; a cut is generated using the best dual multipliers at the time when the issues are detected
  • sub, subgr_stalling: were relevant in an earlier implementation of a subgradient method; not relevant anymore
  • mn_opt: a second optimization step was used (``minimal norm choice''); this step terminated with optimality
  • mn_iter: a second optimization step was used (``minimal norm choice''); this step terminated with the iteration limit
  • mn_issue: a second optimization step was used (``minimal norm choice''); this step terminated with numerical issues
• A section containing the results of an in-sample or out-of-sample simulation conducted after SDDiP has terminated. In each case the information contains
  • the deterministic lower bound
  • the simulated upper bound and a confidence interval

This page was generated using Literate.jl.