• Report.pdf - The report of the project.

• Presentation.pdf - The presentation of the project.

• requirements.txt - Requirements.

• datasets - Folder that includes the dataset for the SAT3 Problem and the dataset for the Graph-Coloring Problem.

• classification/GNN - Folder that includes all prerequisites for the GTN training-tuning-evaluation. More specifically,

  • data_loader.py, data loader that processes raw .cnf data for the SAT3 problem,
  • dataset_pytorch.py, data loader that translates the processed data to the torch.geometric format,
  • model.py, where the GTN model architecture is located,
  • tuning.py, where functions for hyperparameter tuning are located,
  • train.py, where functions for training, testing and logging are located,
  • best_parameters_same_sets.txt & best_parameters_diff_test.txt, where best parameters are stored after tuning (for same distribution sets and different distribution sets, respectively),
  • final_model_same_sets.pth,final_model_diff_sets.pth & final_model_same_sets_c.pth, where best models are stored after tuning (for same distribution sets, different distribution sets and graph coloring, respectively),
  • folder >plots, where logging plots are located,
  • demo.py, demo that showcases the performance of best model (when trained and tested on data from same distribution) on the unseen testing example demo.cnf,
  • GNN/classification.ipynb, a Python notebook where the model is trained, tuned and evaluated (on two different evaluation sets).

• classification/LSTM - Folder that includes all prerequisites for the LSTM training-tuning-evaluation. More specifically,

  • data_loader.py, data loader that processes raw .cnf data for the SAT3 problem,
  • dataset_pytorch.py, data loader that translates the processed data to the torch.utils.data format,
  • model.py, where the LSTM model architecture is located,
  • tuning.py, where functions for hyperparameter tuning are located,
  • train.py, where functions for training, testing and logging are located,
  • best_parameters_same_sets.txt, where best parameters are stored after tuning,
  • final_model.pth, where best model is stored after tuning,
  • folder >plots, where logging plots are located,
  • LSTM/classification.ipynb, a Python notebook where the model is trained, tuned and evaluated.

• graph coloring - Folder that includes all prerequisites for transfer learning the best GTN model for the SAT3 problem to the Graph coloring problem. More specifically,

  • data_loader_c.py, data loader that processes raw .cnf data for the 3-colorable problem,
  • transfer_model.py, where functions for re-training and hyperparameter tuning are located,
  • best_parameters_same_sets.txt, where best parameters are stored after tuning the learning parameters of the transferred model,
  • folder >plots, where logging plots are located,
  • graph_coloring/test_model.ipynb, a Python notebook where the model is re-trained and evaluated.

In order to run the whole process you should execute the notebooks,

1. classification/GNN/classification.ipynb
2. classification/LSTM/classification.ipynb
3. graph_coloring/test_model.ipynb

with the given order.

However, every notebook can also be executed separately.

Otherwise, you could just run the demo. Note that the demo is only for the best-performing model: the one trained and tested on data from the same distribution of SAT3 problem instances.