liamsbhoo/timee-multivariate

TIMEE — Multivariate (beta)

Multivariate checkpoint for TIMEE, fine-tuned for attention-based variate pooling. Each variate is encoded through the shared univariate encoder, and the per-variate representations are fused by a learned attention pool (variate_attn_pool) before the in-context phase — modeling channels jointly rather than independently.

Beta: TIMEE is trained and evaluated as a univariate classifier. Multivariate is not its focus (yet) — this checkpoint is provided so people can use and evaluate it. The main univariate model lives at liamsbhoo/timee.

Usage

from timee import TimeeMultivariateClassifier

clf = TimeeMultivariateClassifier.from_pretrained("liamsbhoo/timee-multivariate")

# X: (n_samples, n_channels, seq_len) float32
predictions, probabilities = clf.predict(X_train, y_train, X_test)

For zero-shot multivariate classification without this checkpoint, TimeeClassifier handles n_channels > 1 by classifying each channel independently and averaging the per-channel class probabilities.

Citation

@misc{küken2026timeeendtoendtimeseries,
      title={TimEE: End-to-end Time Series Classification via In-Context Learning},
      author={Jaris Küken and Shi Bin Hoo and Martin Mráz and Frank Hutter and Lennart Purucker},
      year={2026},
      eprint={2607.07500},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2607.07500},
}