This project is about finding a solution to the traveling-salesman problem (TSP) using a so called goal-based AI agent. The goal is to find a cycle (a roundtrip) which visits every city once, while traveling the minimal possible distance.
A search algorithm called first-choice hill-climbing search has been used, which is a algorithms from the family of local search algorithms. This search evaluates and modifies one current state rather than systematically exploring paths from an initial state to a goal state, as it is done in classical search. This path to reach the goal is not of interest. Thus, no data structure representing the search space needs to be maintained. Only the current node needs to record the state and the value of the objective function, here the distance.
The current sate is a sequence of cities along with the distance. The search creates new randomly sequences and checks if the new sequences has a shorter distance compared to the current sequence. As soon as any shorter distance is found, take this as the new current state and repeat. It resets the best sequence five times in total - every time after 2000 iterations. Hence it checks overall 10,000 sequences. The best sequence within all is the solution and is printed out to the command line in the end.
An alternative version of the hill-climbing algorithm being used is called the steepest-ascent version.
- Install dependencies using
pip install -r requirements.txt - Using Python 3 execute
Hill_climbing_simple.pywith a text file containing the cities with their coordinates likepython3 Hill_climbing_simple.py --filename data/49_cities.txtassuming you are in thesrcdirectory.
- Fully observable
- Single agent
- Stochastic
- Episodic
- Static
- Discrete
- Known
- The final solution is printed at the end of a
log.*.txtfile in theresultsfolder. - Also a plot diagram will be created after search which visualizes the results of the algorithm.
It's storeed in
results/plots/diagramm.*.png.