The tspy package gives a Python framework in which to study the famous Traveling Salesman Problem (TSP). In this package, one can work on specific instances of the TSP. Approximation methods and lower bounds are included by default. The structure of the package makes it easy to create and include your own methods.

The tspy package can be installed using pip:

pip install tspy

To create an instance of the TSP, use:

from tspy import TSP
tsp = TSP()

Currently, data can be given to the instance in two ways: by giving it the NxN distance matrix D or, in the case of an Euclidian TSP, a NxP data matrix X and a distance function.

# Using the distance matrix
tsp.read_mat(D)

# Using the data matrix and a distance function
tsp.read_data(X, dist)

# Using a list of coordinates (with the Euclidean distance)
tsp.read_data([(0,0), (0,1), (1,0), (1,1)])

The module tsp.solvers contains several algorithms providing approximate solutions of TSP instances. For example, the 2-opt algorithm gives good solutions rather quickly. Here is how it can be used in the tspy package:

from tspy.solvers import TwoOpt_solver
two_opt = TwoOpt_solver(initial_tour='NN', iter_num=100)
two_opt_tour = tsp.get_approx_solution(two_opt)

Other solvers are used similarly.

Current solutions are stored in the dictionary tsp.tours. If a data matrix has been provided to the instance, a plot of the solution can be shown:

tsp.plot_solution('TwoOpt_solver')

At any point, the best solution that has been computed can be retrieved:

best_tour = tsp.get_best_solution()

The TSP being NP-hard, it is difficult to get exact solutions for large instances. However, by computing lower bounds we can know how good our approximations are. The tspy package provides several lower bounds methods. One example is given by the Simple_LP_bound which gives the optimal solution of the LP relaxation of the TSP:

from tspy.lower_bounds import Simple_LP_bound
lower_bound = tsp.get_lower_bound(Simple_LP_bound())

At any point, the best lower bound that has been computed can be retrieved:

best_lower_bound = tsp.get_best_lower_bound()

The following features will be added soon:

• Genetic programming
• Ant colony optimization
• Lin–Kernighan heuristic
• Better LP lower bounds

Feel free to contact me if you would like to contribute.

tspy was written by William Borgeaud (williamborgeaud[at]gmail.com).