NRP Solver

Nurse Rostering Problem Solver

Written by Túlio Toffolo (instance reader by Haroldo Santos).
Note that the current code-base employs the formulation from Santos, Toffolo, Gomes and Ribas (2016).

More information: https://benchmark.gent.cs.kuleuven.be/nrp

Please address all contributions, suggestions, and inquiries to the project administrator.

Latest improvements

The current version generates subproblems by employing three decomposition approaches, based on the Time, Nurse and Shift indexes of decision variables X. The subproblems are solved in a random order or following certain priorities. This resulted in quicker convergence than previous algorithms, in addition to improvements over some best known solutions.

Getting Started

The source code includes the decomposition-based heuristic for the NRP.
Note that specific input files are required to execute the solver (available at https://benchmark.gent.cs.kuleuven.be/nrp).

Usage examples:

• ./nurse -dat models/long01.dat -lp models/long01.lp.gz -ini initial_solutions/long01.sol -out output/long01
• ./nurse -dat models/long01.dat -lp models/long01.lp.gz -ini initial_solutions/long01.sol -out output/long01 -lb 197

Requirements

CPLEX 12.7 or Gurobi 7.5

Arguments

Usage: ./nurse <arguments> [options]

Program arguments:
    -dat <file.dat> : nurse problem
    -lp <file.lp>   : nurse problem (in cplex lp format)
    -ini <file.sol> : provide an initial solution
    -out <file.txt> : prefix of files where the solutions will be saved

Optional parameters (example):
    -rseed <int>        : random seed (default: 9)
    -mip_emphasis <int> : mip emphasis (only cplex; default: 2).
                             0   Balance optimality and feasibility
                             1   Emphasize feasibility over optimality
                             2   Emphasize optimality over feasibility
                             3   Emphasize moving best bound
                             4   Emphasize hidden feasible solutions
    -mip_focus <int>    : mip focus (only gurobi; default: 1).
                             0   Balance optimality and feasibility
                             1   Focus on feasibility over optimality
                             2   Focus on optimality over feasibility
                             3   Focus on improving the best bound
    -threads <int>      : number of thread the solver may use (default: 0).
    -time_limit <int>   : runtime limit (default: unlimited).
    -ltime_limit <int>  : solver runtime limit to solve each subproblem  (default: unlimited).
    -iter_limit <int>   : maximum number of iterations (default: unlimited).
    -n_days <int>       : initial number of days in a "time-based" sub-problem (default: 4).
    -days_step <int>    : value of the increment on n_days for the "time-based" decomposition. (default: 3).
    -max_ndays <int>    : maximum value for parameter ndays (default: 30).
    -n_nurses <int>     : initial number of nurses in a "nurse-based" sub-problem. (default: 2).
    -nurses_step <int>  : value of the increment on n_nurses for the "nurse-based" decomposition (default: 4).
    -max_nnurses <int>  : maximum value for parameter n_nurses (default: 5).".
    -lb <int>           : value of the best known lower bound or global optimum (default: 0).

Special parameters:
    --validate-only     : if this parameter is set, the program will only evaluate the solution.

Questions?

If you have any questions, please feel free to contact me.

Thanks!