DiffCATS

Code for "DiffCATS: Causally Associated Time-Series Generation through Diffusion Models".

Create the environment

conda create --name DiffCATS python=3.12.3
conda activate DiffCATS
pip install -r requirements.txt

Run training

In folder src/conf there are the yaml configuration files to set all the parameters of the experiment. Specifically, in src/conf/logger/default.yaml it is possible to configure the wandb logger to track metrics during training. To run an experiment execute the following command:

python src/run.py

An ID associated with each experiment will be generated. Model checkpoints and synthetic generated at the end of each epoch will be saved in storage/ID.

Run inference

To run the generation of synthetic samples execute the following command:

python src/run.py seed=SEED 'ckpt_path="storage/ID/checkpoints/epoch=EPOCH.ckpt"'

Synthetic data will be saved in storage/ID/inference_data.

Run evaluation

To evaluate the synthetic data, e.g. concerning authenticity, execute the following command:

python src/run_authenticity.py +run_id=ID +epoch=EPOCH

For the evaluation through dimensionality reduction techniques execute the following command:

python src/run_tsne_pca_kde.py +run_id=ID +epoch=EPOCH +title=PLOT_TITLE