/atsp-approximation

Reference implementation of the algorithms presented in the paper 'From Symmetry to Asymmetry: Generalizing TSP Approximations by Parametrization'

Primary LanguagePythonMIT LicenseMIT

From Symmetry to Asymmetry: Generalizing TSP Approximations by Parametrization

Build Status

This repository provides a reference implementation of the algorithms presented in the paper 'From Symmetry to Asymmetry: Generalizing TSP Approximations by Parametrization'.

Dependencies

Python >= 3.6 and Java >= 8 are required to run the program. The remaining dependencies can either be downloaded and built automatically by Make (Linux only), or manually. The build process requires a C++ compiler, CMake and Make.

Makefile

Install Poetry, which will manage the Python dependencies:

$ python3 -m pip install poetry --user

Use Make to automatically download and build all dependencies:

$ make all

Manual build

  • Download Concorde and save the binary under vendor/concorde.
  • Either install the LEMON graph library via a package manager, or build it from source.
  • Build the MSA solver by running g++ -O3 lib/msa/msa.cpp -o lib/msa/msa_solver. If you built LEMON from source, you will need to point g++ toward the LEMON headers via the -I flag.
  • Install Poetry by running python3 -m pip install poetry --user. Poetry will manage the Python dependencies.
  • Install the Python dependencies in a new virtual environment by running poetry install.
  • Download the TSPLIB ATSP instances and place them in the instances/tsplib directory.

Running

Enter the virtual environment and execute the main script:

$ poetry shell
$ python3 src/main.py

The script supports the following arguments:

usage: [-h] (-t | -c) [--only-kernel-size | --tour] [-b BETA | --multibeta] graph

positional arguments:
  graph                 A file describing a graph. Multiple formats are supported, identified by the
                        file name extension:
                          .atsp: TSPLIB files with EDGE_WEIGHT_FORMAT=FULL_MATRIX
                          .csv: weight matrices in CSV format
                          .tsv: weight matrices in TSV format
                          .txt: the graph's dimension followed by whitespace-separated edge weights.

optional arguments:
  -h, --help            show this help message and exit
  -t, --treedoubling    Use the generalized tree doubling algorithm
  -c, --christofides    Use the generalized Christofides algorithm
  --only-kernel-size    Only output the instance's kernel size without computing a tour
  --tour                Output the computed tour as a space-separated node list
  -b BETA, --beta BETA  Asymmetry factor above which edges are treated as asymmetric (default: 1).
                        Choosing beta = 0 will compute an exact solution. You may also pass a
                        comma-separated list of values to execute the algorithm multiple times.
  --multibeta           Execute the script multiple times with different values for beta.
                        First, compute an exact solution as a reference point. After that, start
                        by treating every asymmetric edge as asymmetric (beta = 1), then halve
                        the number of asymmetric edges each time until no asymmetric edges remain.