elkai - a Python library for solving TSP problems

• based on LKH by Keld Helsgaun: with proven optimal solutions up to N=315 and more accurate results than Google's OR tools
• asymmetric and symmetric travelling salesman problems support
• clean and simple API: get results with one line calls

Installation

💾 To install it run pip install elkai

Example usage

import elkai

cities = elkai.Coordinates2D({
    'city1': (0, 0),
    'city2': (0, 4),
    'city3': (5, 0)
})

print(cities.solve_tsp()) # Output: ['city1', 'city2', 'city3', 'city1']

import elkai

cities = elkai.DistanceMatrix([
    [0, 4, 0],
    [0, 0, 5],
    [0, 0, 0]
])

print(cities.solve_tsp()) # Output: [0, 2, 1, 0]

Note

solve_int_matrix and solve_float_matrix are deprecated in v1. Also, they don't contain the departure to origin in the result by default.

License

The LKH native code by Helsgaun is released for non-commercial use only. Therefore the same restriction applies to elkai, which is explained in the LICENSE file.

How it works internally

• We refactored LKH such that it doesn't have global state and you don't need to restart the program in order to run another input problem
• We added a hook in ReadProblem that allows reading problems from memory instead of files
• We read the solution from the Tour variable and put it in a PyObject (Python list).
• ✓ Valgrind passed on d3d8c12.

⚠️ elkai takes the global interpreter lock (GIL) during the solving phase which means two threads cannot solve problems at the same time. If you want to run other workloads at the same time, you have to run another process - for example by using the multiprocessing module.

If there isn't a prebuilt wheel for your platform, you'll have to follow the scikit-build process.