Skip to content

README.md

Latest commit

 

History

History
46 lines (34 loc) · 1.71 KB

README.md

File metadata and controls

46 lines (34 loc) · 1.71 KB

dsPUcopula

dsPUcopula provides the server-side machinery that powers the PU copula based synthetic data generator for DataSHIELD. The package wraps the preprocessing, modelling and disclosure control checks that need to run alongside the standard DataSHIELD aggregation interface. In addition to the R components, the package exposes convenience wrappers around the Python packages syndat and anonymeter so that privacy risk metrics can be evaluated directly from the server session.

Installation

The package is not yet on CRAN. You can install the development version from a Git remote (for example GitHub) using:

remotes::install_github("bips-hb/dsPUcopula")

Once installed on the DataSHIELD server you can set up the required Python environment by running:

dsPUcopula::setup_python()

ToDo

  • Introduce a privacy filter option that can force inference_check = TRUE, singling_out_check = TRUE and inference_check_ignore_na = FALSE in generateSyntheticDS.

Overview of the server workflow

The typical order of operations initiated by the DataSHIELD client is:

  1. save_original_varnamesDS() and save_original_classesDS() to capture the metadata of the training data set.
  2. preprocessDataDS() to encode categorical variables while tracking their original levels.
  3. fitPUcopulaDS() and estimateMarginalsDS() to fit the copula and marginal models.
  4. generateSyntheticDS() to draw synthetic data, optionally requesting privacy and utility scores via py_syndat_scores(), py_anonymeter_SinglingOut() and py_anonymeter_Inference().

See the function documentation for full details on the available arguments and customisation options.