PyKEEN Negative Sampling Extension documentation

PyKEEN Negative Sampler Extension from “d’Amato, C., Diliso, I., Fanizzi, N., & Saeed, Z. (2025). Enhancing PyKEEN with Multiple Negative Sampling Solutions for Knowledge Graph Embedding Models (Version 0.0.0.1) [Computer software]. https://doi.org/10.5281/zenodo.15413075”.

This library is a modular and extensible add-on to the PyKEEN framework, designed to provide standardized implementations of advanced negative sampling strategies for knowledge graph embedding (KGE) models.This library introduces both static and dynamic negative samplers—ranging from type-based and relational filters to model-guided adversarial methods—within a unified interface fully compatible with PyKEEN’s training, evaluation, and hyperparameter tuning workflows. Built for researchers and practitioners, the extension simplifies the development and integration of new samplers while ensuring reproducibility and performance diagnostics through automated pool statistics and flexible fallback strategies. Whether you’re testing new sampling techniques or aiming for fair benchmarking across KGE models, this library provides a principled foundation for scalable and customizable negative sampling.

Contents:

• Quick Start Guide
  • Installing requirements
  • Import the extension and setting paths
  • Loading a Dataset
  • Using Static Negative Samplers
  • Using Dynamic Negative Samplers
  • Using Negative Samplers in PyKEEN Pipelines
  • Writing you own custom Negative Samplers
  • Integration of random negatives
  • Generating Negative Sampler Statistics
• PyKEEN Extension
  • Constants
  • Dataset
    • OnMemoryDataset
  • Filtering
    • NullPythonSetFilterer
  • Sampling
    • ClassesNegativeSampler
    • CorruptNegativeSampler
    • NearMissNegativeSampler
    • NearestNeighbourNegativeSampler
    • RelationalNegativeSampler
    • SubSetNegativeSampler
    • TypedNegativeSampler