Credits

Author

Aleksandra Ciprijanovic GitHub: @AleksCipri

ShiftKit is a modular, science-focused domain adaptation framework for PyTorch, built around clean interfaces for fast implementation of different deep learning architectures, datasets, and domain adaptation methods.

---

Other Contributors

Karla Tame-Narvaez GitHub: @karlaTame

Abdelrahman Helal GitHub: @abdelrahman-helal

Added PyTorch Geometric support: GNN model, node-level and graph-level domain adaptation, PyG data utilities, and the node-level MMD example.

---

Dependencies

ShiftKit builds on the following open-source libraries:

Library	Version	Use
PyTorch	≥ 2.0	Deep learning backend
torchvision	≥ 0.15	Built-in datasets (MNIST)
NumPy	≥ 1.24	Numerical operations
scikit-learn	≥ 1.2	t-SNE for latent space visualisation
matplotlib	≥ 3.7	All plotting and diagnostics
tqdm	≥ 4.65	Training progress bars

Optional dependencies (required only for specific methods):

Library	Version	Use
torch-geometric	≥ 2.0	GNN models and PyG graph data (shiftkit.models.GNN)
geomloss	≥ 0.2	Sinkhorn divergence for SIDDATrainer

Documentation built with MkDocs and the Material for MkDocs theme.

---

Licence

ShiftKit is released under the Apache License 2.0.