TIMEE: Time Series Classification via In-Context Learning

TIMEE is a pretrained transformer for time series classification, introduced in TimEE: End-to-end Time Series Classification via In-Context Learning.

It classifies test series in a single forward pass given labeled training examples — no per-dataset training or fine-tuning required.

Official GitHub Repository: https://github.com/automl/timee

Usage

pip install timee-ts

from timee import TimeeClassifier
import numpy as np

# Downloads weights automatically on first use
clf = TimeeClassifier.from_pretrained()

# X: (n_samples, n_channels, seq_len) float32
X_train = np.random.randn(20, 1, 256).astype(np.float32)
y_train = np.array([0, 1] * 10)
X_test  = np.random.randn(5,  1, 256).astype(np.float32)

predictions, probabilities = clf.predict(X_train, y_train, X_test)

Labels can be any type (int, str, etc.). Datasets with more than 10 classes are handled automatically via one-vs-rest.

UCR Benchmark Results

Dataset	Classes	Accuracy	ROC AUC
ArrowHead	3	76.6 %	0.964
ECG5000	5	95.0 %	0.952
GunPoint	2	98.7 %	0.997
ItalyPowerDemand	2	96.0 %	0.993
TwoPatterns	4	99.8 %	1.000

Results use the default 4-member ensemble (interpolate × {256, 512} × {raw, first-difference}).

Model Details

• Parameters: 4,557,322
• Input: univariate or multivariate time series, any length
• Output: class probabilities over up to 10 classes (OvR for more)
• Inference: single forward pass; no dataset-specific adaptation

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},
}