This is an implementation of the new neighborhood searches for one-to-one pickup-and-delivery problems. More precisely, we focus on the pickup-and-delivery traveling salesman problem (PDTSP), which is the canonical problem in this family [1].
Main project features:
- Relocate-Pair neighborhood: Enhanced implementation in O(n2) instead of O(n3) time.
- 2-Opt neighborhood: Adaptation to the PDTSP.
- Or-Opt neighborhood: Adaptation to the PDTSP.
- 2k-Opt neighborhood: New exponential-size neighborhood containing all solutions formed with nested 2-Opts and explored in O(n2) time.
- 4-Opt neighborhood: Based on the concepts of Glover (1996) and adapted to the PDTSP.
- Balas and Simonetti neighborhood: Based on the concepts of Balas and Simonetti (2001) and adapted to the PDTSP.
- All these neighborhoods are integrated within the hybrid genetic search (HGS) of Vidal et al. (2012, 2014).
- We also provide a reimplementation of Veenstra et al. (2017) Ruin and Recreate algorithm, including an O(n2) reinsertion operator.
When using this algorithm (or part of it) in derived academic studies, please refer to the following work:
[1] Pacheco, T., Martinelli, R., Subramanian, A., Toffolo, T. A., & Vidal, T. (2021). Exponential-Size Neighborhoods for the Pickup-and-Delivery Traveling Salesman Problem. Transportation Science, Forthcoming. ArXiv preprint arXiv:2107.05189.
This code has been designed to solve the Pickup-and-Delivery Traveling Salesman Problem (PDTSP).
This code version has been designed and calibrated for medium-scale instances with up to 1,000 customers.
You need CMake
to compile.
Build with:
mkdir build
cd build
cmake ..
make -j4
This will generate the executable file pdphgs
and pdprr
in the build
directory.
After building the executables, you can try an example of pdphgs
:
./pdphgs --instance=../instances/RBO00/Class1/U159C.PDT --it=1000
The following options are supported: (default values are included in parentheses)
./pdphgs --help
--help Display a help message.
--version Display the current version.
--verbose Enable verbose mode.
--instance Instance file path.
--grubhub Read file as a distance matrix (GrubHub
format).
--seed arg (=0) Sets a random seed.
--time-limit arg (=2147483647) Set the maximum execution time in seconds.
--bs-k arg (=3) Balas&Simonetti k parameter.
--or-k arg (=30) Or-Opt k parameter.
--ratio-slow-nb arg (=1) Ratio of slow neigborhoods usage in local searches.
--neighborhoods arg (=RELOCATE-2OPT-2KOPT-OROPT-4OPT-BS)
Select neighborhood structure.
--mu arg (=25) Minimum population size.
--it arg (=1000000) Maximum iterations without improvement.
--div arg (=4000) Iterations without improvement that trigger. diversification
--lambda arg (=40) Number of offspring in each generation.
--nb-elite arg (=1) Number of elite individuals.
--nb-close arg (=2) Number of close individuals.
You can also run the Ruin and Recreate implementation as follows:
./pdprr --instance=../instances/RBO00/Class1/U159C.PDT --fast
The following options are supported:
./pdprr --help
--help Display a help message.
--version Display the current version.
--instance arg Instance file path.
--fast Use the fast reinsertion operator.
--seed arg (=0) Seed provided to srand.
--p-accept arg (=3) p parameter to accept a request as worst.
--c-rate arg (=0.99987571600000003) Cooling rate.
--it arg (=50000) Maximum number of iterations.
--time-limit arg Set maximum execution time in seconds.
The main folders containing the logic of the algorithm are the following:
- hgsadc: Implementation of HGS with a generic problem and solution interface.
- pdp: Implementation of the local search and other operators required by the GA (crossover, repair and mutate).
- pdp/moves: Generic interface for moves (PDPMoves), and individual classes with the implementations of the six neighborhoods: Relocate-Pair, 2-Opt, Or-Opt, 2k-Opt, Balas and Simonetti and 4-Opt.
In addition, additional classes have been created to facilitate interfacing:
- Application: Reads command line and stores the parameters of the algorithm.
- random: Utilities to initiate RNG and allow ranged sampling.
- main: Main code to start the algorithm
The main classes containing the logic of the algorithm are the following:
- Application: Reads command line and stores the parameters of the algorithm.
- Instance: Reads instance file and stores precompute distances.
- Operators: Implementation of the Removal and Reinsertion (classical and fast) operators.
- Solution: Represents an individual solution.
- Solver: Implementation of the Ruin and Recreate algorithm.
- instances/RBO00: Folder with instances introduced in Renaud et al. (2000) - (.sol files containing the best-known solutions found).
- instances/Dumitrescu: Folder with instances introduced in Dumitrescu et al. (2010) - (.sol files containing the best-known solutions found).
- instances/Grubhub: Folder with instances introduced in O’Neil (2018).
- instances/PDP-X: Folder with instances introduced in Pacheco et al. (2022) - (with .sol files containing the best-known solutions found).
Thank you very much for your interest in this code. This code is actively maintained and evolving. Pull requests and contributions seeking to improve the code in terms of readability, usability, and performance are welcome. Development is conducted in the dev branch. I recommend contacting us beforehand at [email protected] or [email protected] before any major rework.
- MIT license
- Copyright(c) 2022 Toni Pacheco