This is the release of my implementation of the PDSTSP solver (Parallel Drone Scheduling Traveling Salesman Problem), as stated in the paper mentionned in the README.md file.
There are two files:
- The "_IntCallback" implements Lazy Callback, which means that the SEC (Subtour Elimination Constraint) is added continuously, as soon as the solver finds a solution, the Lazy Callback is called and it'll check if the SECs are respected for that specific solution. This workflow allows for faster iteration (in my tests with n=15, it was solved in around 6 seconds).
- The "_NoCallback" applies the basic mathematical model, and thus calculates all the SECs at start (the SECs are exponential: 2^n). This works pretty well but it takes a significant amount of time (in my tests with n=15, it was solved in around 2 minutes).
One thing important to note, I have not implemented the User Callback, which takes fractionnal solutions before the solver even finds a feasible solution to report to the Lazy Callback. This is the cause, in my opinion, of a difference in the results between these two implementations (both results were correct, but the second was not the ideal one).