Reproducible code for SOLMformer, a metadata-aware categorical time series generative model.
SOLMformer/
├── README.md
├── requirements.txt
├── data/
├── models/
│ ├── __init__.py
│ ├── gpt.py
│ └── model.py
├── train/
│ ├── __init__.py
│ ├── train_baseline.py
│ └── train_solmformer.py
└── utils/
├── __init__.py
├── BPIDataset.py
└── Evaluator.py
export PYTHONPATH="$PWD:${PYTHONPATH}"if you get issues with importing items from one directory into another.pip install -r requirements.txtto install dependencies.
You will need to download datasets
- BPI Challenge 2020i: International Declarations
- BPI Challenge 2017
- BPI Challenge 2015 - 1
- BPI Challenge 2013
- BPI Challenge 2012
After downloading, unzip and put the .xes file in the data directory.
The following train/eval instructions assume that the file data/{dataset-name}.xes exists.
Run the command
python train/baseline.py --dataset-name={dataset-name}
python train/solmformer.py
--logits_weight=1.0 --durations_weight=0.0 --dataset-name={dataset-name}
python train/solmformer.py --logits_weight=0.0 --durations_weight=1.0 --dataset-name={dataset-name}
python train/solmformer.py --logits_weight=0.0 --durations_weight=0.0 --dataset-name={dataset-name}
python train/solmformer.py
--logits_weight={w1}
--durations_weight={w2}
--dataset-name={dataset-name}