Solving TSP with GNN - Training Attention Model

The code of the project can be found in 'code' folder.
The file Attention has the attention based TSP solver.
Train Policy file has the training procedure of the code.
The parameter file uploaded is a trained model parameter for TSP50.
TSSP50 file has approximately 1200 random TSP graphs and its solutions. We used these instances to test the model performance.
The simple_TSP file is used to test the original attention model's rotation invariance.
You may check the training and model files however, the only results we provide is from simple_TSP file.
You may run the simple_tsp jupyter notebook it will generate the graphs given in the report.