Skip to content
dieterich-labPublic

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

21 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ASyH-RL

The data synthesis tool based on principles of Q-learning for h-parameters optimization

Overview

The fork of ASyH, enhanced with a module that runs a reinforcement learning optimization of generative model hyperparameters. The RL-agent runs a closed optimization loop for predefined number of episodes, adjusting model parameters at the end of each iteration. Its main stages follow the order: generation -> evaluation -> adjustment of parameters .

At this moment there are 5 available models in the toolkit, although we plan to extend the list:

ID Model Description
0 CTGAN Conditional tabular generative adversarial network
1 CopulaGAN Copula generative adversarial network
2 TVAE Tabular variational autoencoder
3 Gaussian Copula —
4 Gaussian Mixture —

Required dependencies & libraries

Necessary libraries are specified in the file requirements.txt . To install needed dependencies run in cmd inside the right virtual environment: $ pip install -r requirements.txt

How to use

To launch the tool you can use the sbatch script launch_sbatch_model.sh - on HPC it submits a job to SLURM, that runs an RL-optimization routine for a model, whose number is provided as an argument; for example $ sbatch launch_sbatch_model.sh 0 - it will start a process for CTGAN & RL agent, depending on what model you need replace the digit with another one based on the table above in the overview section.

Essential files and folders

  • launch_sbatch_model.sh - on HPC it submits a job to SLURM, that runs an RL-optimization routine for a model, whose number is provided as an argument; for example $ sbatch launch_sbatch_model.sh 0 - it will start a process for CTGAN & RL agent

  • main.py - the main python script, it executes parameter optimization, model fitting and data synthesis

  • rl_agent/ - the folder contains the module with the routines for RL agent

  • evaluator/utility_evaluator_2.py - responsible for calculating utility score of produced data

  • evaluator/fidelity_evaluator.py - for calculating fidelity of synth. data based on statistical tests (KS, JS, TVD)

  • rl_optimizer.py - the implementation of closed optimization loop

About ASyH (the original toolkit)

A data protection tool that utilizes generative machine learning models to create synthetic datasets, safeguarding sensitive patient information. The ASyH is a software helping Clinics as holders of large quantities of highly restricted personal health data to provide the Medical Data Community with realistic datasets without the breach of privacy. It does this by synthesizing data with Machine Learning techniques which preserve data distribution and correlation while adding as much variation to the synthetic data, in order for it to have no resemblance to any of the original patient data entries. For synthesis, metrics and quality assurance we will mainly use the Synthetic Data Vault (github).

More info can be found here - https://github.com/dieterich-lab/ASyH/blob/main/README.md.

References

  1. Johann, T. I., Otte, K., Prasser, F., & Dieterich, C. (2025). Anonymize or synthesize? Privacy-preserving methods for heart failure score analytics. European Heart Journal-Digital Health, 6(1), 147-154.
  2. Giomi, M., Boenisch, F., Wehmeyer, C., & Tasnádi, B. (2022). A unified framework for quantifying privacy risk in synthetic data. arXiv preprint arXiv:2211.10459.
  3. Wang, S., Wang, J., Cheng, X., Zhang, W., Tian, Y., Ning, M., ... & Liu, Y. (2026). CLEO closed loop framework for synthesizing medical privacy preserving tabular data. npj Digital Medicine.
  4. Montanez, A. (2018). SDV: an open source library for synthetic data generation (Doctoral dissertation, Massachusetts Institute of Technology).

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages