pykeen

Why we included this project

For anyone doing link prediction or knowledge base completion, PyKEEN bundles roughly forty knowledge graph embedding models behind one consistent API, so you can compare approaches like TransE and ComplEx without reimplementing them. Training a model on a built-in dataset is essentially a single pipeline call, and the package ships dozens of benchmark datasets, including inductive splits and large biomedical graphs. It integrates with Optuna for hyperparameter search and runs on PyTorch Lightning and CUDA, so it slots into existing deep learning workflows. Teams that need to fill missing triples or reason over entity relationships get a direct path from raw triples to evaluated predictions, with most boilerplate removed.