Installation

Requirements

• Python 3.9 or newer
• pip (or another PEP 517–compatible installer)

---

Install via pip

pip install fairness_training[train,verify]

This installs the core package with PyTorch and the CVXPY fairness layer — the typical setup for training.

Install options

Extra	What it adds	When to use
train	PyTorch, torchvision, NumPy, SciPy, Transformers	Model training
verify	CVXPY, cvxpylayers, ECOS, diffcp	Fairness layer — required for FairModel
full	Everything above + datasets, vision tools, TensorBoard	All features
cpu	CPU-only PyTorch + NumPy	CPU-only environments (pair with verify: [cpu,verify])
viz	Matplotlib	Plotting utilities (fairness_training.viz)
dev	pytest, pytest-cov, black	Running the test suite

# Training + fairness + visualization
pip install fairness_training[train,verify,viz]

# CPU-only (no CUDA) + fairness layer
pip install fairness_training[cpu,verify]

# Full installation
pip install fairness_training[full]

# Development
pip install fairness_training[dev]

---

Install from Source

For the latest development version:

git clone https://github.com/dtroxell19/fairness_training.git
cd fairness_training
pip install -e .[train,verify]

---

Verify Installation

import fairness_training
print(fairness_training.__version__)  # 0.1.0

---

Troubleshooting

AttributeError: module 'onnxscript.values' has no attribute 'ParamSchema'

A conflict with onnxscript, which this library doesn't use:

pip uninstall onnxscript -y

Solver issues

The verify extra installs ECOS, which handles most problems. If you hit numerical issues, try:

pip install clarabel   # recommended alternative
pip install scs        # another option

---

Next Steps

Head to the Quickstart to train your first fairness-aware model.