charon25 / genetictspsolver

Genetic algorithm to find an approximate solution of the Travelling Salesman Problem.

Genetic Travelling Salesman Problem solver

This repo contains a genetic algorithm (GA) designed to solve the Travelling Salesman Problem (or TSP).

It works by repeatedly generating new paths from previous one in an attempt to improve them.

Install

To install the package :

git clone https://github.com/charon25/GeneticTSPSolver.git
cd GeneticTSPSolver
python setup.py install --user

This package requires numpy (>= 1.21.2) and tqdm (>= 4.62.3).

Usage

To use the solver, just import the package and instantiate the class :

from gentsp import TSPSolver

solver = TSPSolver(distances)

solver.pass_generations(n_generations=100)

The solver has only one mandatory argument : a square distances matrix which indicates the distance between every two pair of points. Everything else is either optional or deduced.

Optional keyword arguments

• population_size (int, value > 0, default: 100): size of the population of the GA.
• mutation_rate (float, 1.0 >= value >= 0, default: 0.05): probability of mutation of every individual
• new_individuals (int, value >= 0, default: 0): number of new random individuals in each generation (introduces novelty).
• elitism (int, value >= 0, default: population_size / 5): number of individuals to keep between every generation (only the best are kept).
• selection (str, 'best' or 'weighted', default: 'weighted'): method to chose the individuals which will breed:
  • best will use the best one (sort by increasing distance)
  • weighted will pick randomly with a probability proportional to the inverse of the distance
• breeder_count (int, value > 0, default: population_size / 2): number of individual who will breed.

Example :

solver = TSPSolver(distances, elitism=30, population_size=400, selection='best', breeder_count=30)

Statistics

The function pass_generations return a dictionary with the following statistics (each as a list of the value for each generation) :

• best_fitness, average_fitness, worst_fitness
• best_distance, average_distance, worst_distance
• fitness_stdev: standard deviation of the fitnesses
• best_individual, average_individual (the closest to the average fitness), worst_individual

Examples

With 25 random points (0 <= x < 100, 0 <= y < 100) and the following code :

solver = TSPSolver(distances, elitism=20)

With 50 random points (0 <= x < 100, 0 <= y < 100) and the following code :

solver = TSPSolver(distances, population_size=300, elitism=20, new_individuals=10)