Solving complex real-world COPs with limited data / information and deep learning Here are some details about the project and what one can find in the report or repository.
- Goal - Finding an optimal solution to Travelling Salesman Problem
- TSP Instances - Contains data in form of coordinates as to show the positions of the cities
- Input - Form of x and y coordinates
- Output - Optimal cost or travelling from the starting city to the city at last, Optimal way of cities visited in order as a sequence.
- Small introduction to different Combinatorial problems
- Databases to find datasets for those COPs
- As majority of datasets do not already have optimal solutions (the dataset we use have an optimal solution given) heuristic methods like Two-opt, City-swap, genetic, Simulated Annealing algorithms are taken into account to get a soution in order to compare the results of Pointer Networks with Transformers.
- An explanation on Pointer Networks, Transformers Architecture and how both can be merged in order to get a working algorithm.
- Conclusion - Transformers performed well in the training set but could not generalize well.
[2023]