Skip to content
Python 3 TSP solver based on LKH (cross platform)
C C++ Other
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
LKH-2.0.9 Remove MSVC sln and modify CI May 4, 2019
elkai Add matrix type Oct 20, 2019
.gitignore Indentation fix in _elkai.c May 3, 2019
.travis.yml Bump python versions for Travis Oct 24, 2019
CMakeLists.txt Add PIC May 3, 2019
LICENSE Update LICENSE May 18, 2019 Update Oct 22, 2019
appveyor.yml Make sure PowerShell Core doesn't run on Linux Oct 24, 2019 Reuse CMake from 3.7 in 3.8 Oct 24, 2019

elkai - a Python 3 TSP solver

elkai is a Python 3 library for solving travelling salesman problems without external dependencies, based on LKH by Keld Helsgaun.

💾 To install it run pip install elkai

💻 Supported platforms: elkai is available on Windows, Linux, OS X for Python 3.5 and above as a binary wheel.

Build status image

Example usage

import numpy as np
import elkai

M = np.zeros((3, 3), dtype=int)
M[0, 1] = 4
M[1, 2] = 5
elkai.solve_int_matrix(M) # Output: [0, 2, 1]


solve_int_matrix(matrix, runs=10) solves an instance of the asymmetric TSP problem with integer distances.

  • matrix: an N*N matrix representing an integer distance matrix between cities.

    An example of N=3 city arrangement:

    [                  # cities are zero indexed, d() is distance
        [0, 4,  9],    # d(0, 0), d(0, 1), d(0, 2)
        [4, 0, 10],    # d(1, 0), d(1, 1), ...
        [2, 4,  0]     # ... and so on
  • runs: An integer representing how many iterations the solver should perform. By default, 10.

  • Return value: The tour represented as a list of indices. The indices are zero-indexed and based on the distance matrix order.

solve_float_matrix(matrix, runs=10) has the same signature as the previous, but allows floating point distances. It may be inaccurate.


What's the difference between LKH and elkai?

elkai is a library that contains the LKH solver and has some wrapper code to expose it to Python. The advantage is that you don't have to compile LKH yourself or download its executables and then manually parse the output. Note that elkai and its author are not affiliated with the LKH project. Note: Helsgaun released the LKH project for non-commercial use only, so elkai must be used this way too.

How to manually build the library?

You need CMake, a C compiler and Python 3.5+. You need to install the dev dependencies first: pip install scikit-build ninja. To build the library, run python build and python install to install it. To make a binary wheel, run python bdist_wheel.

How accurately does it solve asymmetric TSP problems?

Instances with known solutions, which are up to N=315 cities, can be solved optimally.

Can you run multiple threads?

The library doesn't not release the GIL during the solution search, so other threads will be blocked. If you want to solve multiple problems concurrently you may try the multiprocessing module.

P.S.: The LKH solver was written in C with one problem per process in mind. We try to clean up the global state before and after solving a problem, but leaks are possible - if you want to help out, run Valgrind or similar diagnostics and submit an issue/PR.

You can’t perform that action at this time.