Bayes Vulnerability for Microdata library
Project description
BVM library
Quantitative Information Flow assessment of vulnerability for microdata datasets using Bayes Vulnerability.
DOI (v1.0): 10.5281/zenodo.6533704.
This repository provides an implementation of the paper Flexible and scalable privacy assessment for very large datasets, with an application to official governmental microdata (DOI: 10.56553/popets-2022-0114, arXiv: 2204.13734) that appeared in PoPETs 2022, and of the masters thesis A formal quantitative study of privacy in the publication of official educational censuses in Brazil (DOI: hdl:1843/38085). Please refer to the folder examples for the Notebooks containing the actual results for the experiments performed.
Installation
Use the package manager pip to install bvmlib
.
pip install bvmlib
Usage
Warning: Please fill NA
and NaN
values!
A fix will be provided in a later version.
Meanwhile, consider using the pandas .fillna()
method before calling the BVM()
class, e.g. when creating the pandas DataFrame, as shown below.
Single-dataset
import pandas
from bvmlib.bvm import BVM
# Create a pandas DataFrame for your data.
# For instance:
df = pandas.read_csv(file.csv).fillna(-1)
# Create an instance.
I = BVM(df)
# Assign quasi-identifying attributes.
I.qids(['attribute_1','attribute_2'])
# Assign sensitive attributes (optional).
I.sensitive(['attribute_2','attribute_3'])
# Perform vulnerability assessment.
I_results = I.assess()
# Print re-identification results.
print(I_results['re_id'])
# Print attribute-inference results (only if computed).
print(I_results['att_inf'])
Additional examples
Please refer to the folder examples for additional usage examples, including attacks on longitudinal collections of datasets.
Note on the results
For privacy assessment of Collective Re-identification (CRS / CRL), for each list of quasi-identifying attributes (QID), the following results are computed:
- dCR: corresponds to the deterministic metric;
- pCR: corresponds to the probabilistic metric;
- Prior: corresponds to the adversary's prior knowledge in a probabilistic attack;
- Posterior: corresponds to the adversary's posterior knowledge in a probabilistic attack;
- Histogram: corresponds to the distribution of individuals according to the chance of re-identification.
For privacy assessment of Collective (sensitive) Attribute-inference (CAS / CAL), for each list of quasi-identifying attributes (QID) and for each sensitive attribute (Sensitive), the following results are computed:
- dCA: corresponds to the deterministic metric;
- pCA: corresponds to the probabilistic metric;
- Prior: corresponds to the adversary's prior knowledge in a probabilistic attack;
- Posterior: corresponds to the adversary's posterior knowledge in a probabilistic attack;
- Histogram: corresponds to the distribution of individuals according to the chance of attribute-inference.
Contributing
Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.
Please make sure to update tests as appropriate.
License
GNU LGPLv3 [^compatibility].
[^compatibility]: To understand how the various GNU licenses are compatible with each other, please refer to:
https://www.gnu.org/licenses/gpl-faq.html#AllCompatibility
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.