DP+Fair Benchmarking Framework
This repository provides a Python framework for benchmarking fairness mechanisms on Differentially Private Synthetic Data.
Features
- ⚡ Simple, reproducible setup for benchmarking algorithms
- 🧩 Flexible API to plug in any classifier implementing
fit,predict, andpredict_proba - 📊 Pre-offered datasets included under
data/ - 🔬 Configurable experiment settings: dataset schema, dataset synthesizer, seeds, privacy-budget, input/outputs, classifier, data pre-processing.
Installation
First, we strongly recommend creating a new Python environment before installing the package. This helps maintain a clean and reproducible setup, facilitates dependency and version management and update, and minimises potential conflicts with previously installed libraries.
python3 -m venv dpfair-env
source dpfair-env/bin/activate # On Windows: dpfair-env\\Scripts\\activate
# Installation approach via PyPi or source installation
When using Python 3.9, a specific version of the MBI library (private-pgm) must be installed by running the following command:
pip install git+https://github.com/ryan112358/private-pgm.git@01f02f17eba440f4e76c1d06fa5ee9eed0bd2bca
If you are a Windows or Linux user, please install PyTorch CPU-only dependencies: [Step not required for Mac users]
pip install torch --index-url https://download.pytorch.org/whl/cpu
Users on MacOS are required to install llvm-openmp, either using conda or brew, as below:
Conda:
conda install -c conda-forge llvm-openmp
Brew:
brew install libomp
Then, the library installation can proceed as per usual.
Our recommended method is via PyPI. With a Python 3.9+ and <3.13 environment:
pip install BenchmarkDPFair
Alternatively, there is the possibility to install from source, which also gives access to the data/, example/, and notebook/ directories (be aware that these may consume more resources on your disk given the data provided):
git clone https://github.com/vinicius-verona/dp-fair-intervention-benchmark.git
cd dp-fair-intervention-benchmark
pip install -e .
Repository Structure
├── data/ # Pre-offered datasets
├── src/ # Core source code
├── example/ # Files used in the main paper and a dummy example.
└── README.md
Quick Start
Here is a dummy example:
import argparse
from typing import List, Union
from BenchmarkDPFair.DataGenerator import generate_data, DatasetGeneratorConfig
from BenchmarkDPFair.Benchmark import benchmark, BenchmarkInfo, BenchmarkDatasetConfig
from sklearn.linear_model import LogisticRegression
ESTIMATOR_PARAMS = {
'max_iter': 10000,
'solver': 'saga',
'l1_ratio': 0.5,
'C': 0.8
}
lr = LogisticRegression
classifiers = [lr]
ckwargs = [
ESTIMATOR_PARAMS,
]
classifier_name = ["LR"]
combinations = [
(0, 0),
(0, 1),
]
synths = ["aim", "mst"]
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Arguments of Data Generation for Adult")
parser.add_argument(
"--seeds", "-s",
nargs="+", # 1 or more values
type=int # convert automatically to int
)
args = parser.parse_args()
seeds = args.seeds
eps : List[Union[int,float]] = [.05, .1]
for synthesizer in synths:
for s in seeds:
data_conf = DatasetGeneratorConfig(
name = "Compas",
target= "two_year_recid",
synthesizer = synthesizer,
root_dir="./data",
sensitive_attr = "race",
categorical_cols = ['race', 'score_text', 'c_charge_degree','age', 'sex', 'two_year_recid'],
sensitive_cols = ['race', 'sex'],
ordinal_cols = ['priors_count'],
privacy_budgets=eps,
binary_encoder=binary_encode,
seed = s,
test_split_size=0.4,
data_filter = filter_compas
)
generate_data(f"compas.csv", "", data_conf, "./data", verbose=True)
for clf_idx, syn_idx in combinations:
classifier = classifiers[clf_idx]
synth = synths[syn_idx]
benchmark_config = BenchmarkInfo(
dp_method=synth,
output_dir=f"./output/Dummy-Compas/{classifier_name[clf_idx]}/",
seeds=seeds,
eps = eps,
classifier=classifier,
classifier_kwargs=ckwargs[clf_idx]
)
benchmark_dataset = BenchmarkDatasetConfig(
name = "Compas",
target= "two_year_recid",
root_dir="./data",
sensitive_attr = "race",
index_col="Unnamed: 0",
categorical_cols = ['race', 'score_text', 'c_charge_degree','age', 'sex', 'two_year_recid'],
ordinal_cols=["priors_count"],
sensitive_cols = ['race', 'sex'],
)
benchmark(benchmark_info=benchmark_config, data_conf=benchmark_dataset)
More detailed examples can be found in the example/ directory.
License
License: MIT
Contributing
Contributions are welcome:
- Open an issue for bug reports or feature requests
Metadata
Release files for BenchmarkDPFair 0.2.7
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| benchmarkdpfair-0.2.7.tar.gz | 24.0 kB | Without provenance | Details for benchmarkdpfair-0.2.7.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| benchmarkdpfair-0.2.7-py3-none-any.whl | Python 3 | none | any | Details for benchmarkdpfair-0.2.7-py3-none-any.whl |
Total release size: 52.0 kB
Release files / benchmarkdpfair-0.2.7.tar.gz
| Download URL | benchmarkdpfair-0.2.7.tar.gz |
|---|---|
| Size | 24.0 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
c94d9b08065c8b651ba197dee6d6c1ca23acc99d74548e425505c6a1067fc2a5
|
|
BLAKE2b-256 checksum How to use checksums |
e74c4d8ae7de2e2758acca60af9d0c6a3c907a219d8f4bb8650c0ac86e1ba25d
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.12.3
|
Release files / benchmarkdpfair-0.2.7-py3-none-any.whl
| Download URL | benchmarkdpfair-0.2.7-py3-none-any.whl |
|---|---|
| Size | 28.0 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
5ac62389f7405c351833fc1fa1241464eb601436a54e26080dd402bb1f2c9c51
|
|
BLAKE2b-256 checksum How to use checksums |
ab422d3cc8c977a67b64e92ca88fe6eaa221bbbe3b90e388d098525679e3c0ac
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.12.3
|
Security for BenchmarkDPFair 0.2.7
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.