Contrastive learning with triplet mining — hard, semi-hard, and easy triplet mining strategies for embedding space optimization. Pure Python.
- Triplet mining strategies — hard, semi-hard, easy, and random mining
- Distance metrics — Euclidean, cosine, and custom distance functions
- Batch mining — efficient mining within mini-batches
- Pure Python — zero external dependencies, works with any embedding format
pip install triplet-minerEmbedding optimization tool for the SuperInstance ecosystem. Used by vector-novelty for improving novelty detection accuracy and by conservation-spectral for spectral embedding quality.
MIT