pykeen: a Python framework for training and evaluating knowledge graph embedding models

What it solves

PyKEEN simplifies the process of training and evaluating knowledge graph embedding models. It provides a standardized framework for researchers and developers to experiment with various embedding techniques, including those that incorporate multi-modal information, without having to implement the models or datasets from scratch.

How it works

PyKEEN uses a high-level pipeline function that allows users to specify a model and a dataset to quickly start training and evaluation. The library is designed to be extensible, featuring a consistent API across different models and training loops (such as the stochastic local closed world assumption), and providing tools like TriplesFactory for handling custom datasets.

Who it’s for

It is designed for AI researchers and developers working with knowledge graphs and embedding models who need a robust, extensible Python package for model training and evaluation.

Highlights

• Includes 37 built-in datasets and 5 inductive datasets.
• Supports 40 different embedding models.
• Integrated with Optuna for hyperparameter optimization and PyTorch Lightning for scalable training.
• Provides a high-level pipeline for rapid prototyping and evaluation.