Installation¶

Requirements¶

Python 3.10+
PyTorch 2.x
macOS (MPS), Linux (CUDA), or CPU

Install from PyPI¶

pip install auditml

Install from source¶

git clone https://github.com/EemanAsghar/AuditML-Privacy-Toolkit.git
cd AuditML-Privacy-Toolkit
pip install -e ".[dev]"

The [dev] extra installs pytest, ruff, and mkdocs tooling.

Optional: Rust acceleration¶

AuditML ships a Rust extension that speeds up threshold scanning (~11× faster) and SSIM computation (~3× faster). It falls back to NumPy automatically if unavailable.

Prerequisites¶

Install Rust (one-time, global):

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
source "$HOME/.cargo/env"

Install maturin inside your virtual environment:
```
pip install maturin
```

Build and install the extension¶

cd rust
maturin build --release --out ../dist
pip install ../dist/auditml_rust-*.whl --force-reinstall

To verify:

from auditml.utils.rust_accel import rust_status
print(rust_status())
# Rust extension available (auditml_rust vX.Y.Z)

Verify the full install¶

python -c "import auditml; print(auditml.__version__)"
auditml --help

Dependencies¶

Package	Purpose
`torch`	Model training and inference
`torchvision`	Datasets and transforms
`opacus`	Differential Privacy
`scikit-learn`	Metrics (AUC, F1, …)
`numpy`	Array operations
`matplotlib`	Plots
`tqdm`	Progress bars
`pyyaml`	Config parsing