A fork of yd-kwon/POMO/NEW_py_ver
We provide codes for two CO (combinatorial optimization) problems:
- Traveling Salesman Problem (TSP)
- Capacitated Vehicle Routing Problem (CVRP)
Other than the re-organized structure, no major change has been made, so that the two versions should give roughly the same results. Some meaningful changes are:
- Query token in the decoder does not contain "graph encoding" now, because this does not seem to make much difference. (But who knows?)
- Normalization methods have changed, from BatchNorm to InstanceNorm. (It seemed more natural. But this may have affected the model's performance in a negative way.)
To train a model, run train.py. train.py contains parameters you can modify. At the moment, it is set to train N=20 problems.
To test a model, run test.py. You can specify the model as a parameter contained in test.py. At the moment, it is set to use the saved model (N=20) we have provided (in result folder), but you can easily use the one you have trained using train.py.