Incorporating Dynamic Graphs into Graph Neural Networks for Business Processes Redesign and Concept Drift Prediction

• Generation of dataset is done using process_dataset_generator.py
• bpmn_dataset.py is the dataset class
  • Note that for now, the total number of graphs in the dataset must be set manually in the dataset during the process model import
• sampling.py allows for faster manual sampling of the edge threshold by saving and then reimporting predicted adjacency matrices
• Model and training loop can be found in
  • previous_iterations/process_gcnn.py
  • previous_iterations/process_vgae.py
  • directed_process_vgae.py + layers.py
    for the three model iterations
• Dataset and logs will be written to ./data-dump folder
• Visualize loss tensorboard by running tensorboard --logdir ./data-dump/logs

Other folders:

• ./literature-review contains the literature corpus and review script
• ./legacy Contains small experiments and old code, neither of which are documented in detail. Included more for completeness than anything else.

In order to ensure that are running the most recent version of the code, check out the GitHub repository here.