TS2Vec
Summary
TS2Vec is a self-supervised time-series representation model built around hierarchical contrastive learning over augmented temporal contexts.
Role In The Wiki
TS2Vec is a pre-TSFM-era representation-learning anchor. Its timestamp-level embeddings, hierarchical contrastive loss, and context-consistency objective are useful reference points when comparing newer masked-reconstruction and foundation-model approaches.
Official Artifacts
- Official code: https://github.com/zhihanyue/ts2vec
- ETNA TS2Vec embedding model: https://docs.etna.ai/2.9.0/api_reference/api/etna.transforms.embeddings.models.TS2VecEmbeddingModel.html
- ETNA pretrained embedding tutorial: https://docs.etna.ai/2.9.0/tutorials/210-embedding_models.html
Evidence
Relation To Foundation TSFM Agenda
Use the source-level agenda mapping in ts2vec-2021 rather than duplicating verdict rows here.
At the entity level, TS2Vec is a pre-TSFM-era representation-learning anchor. Its timestamp-level embeddings, hierarchical contrastive loss, and context-consistency objective are useful reference points when comparing newer masked-reconstruction and foundation-model approaches. This page should stay as the object card; source pages carry slot-level verdicts, evidence, and missing pieces.