aeon

Time series classification applies ML algorithms to sequential data like sensor readings or ECG signals. Aeon provides specialized classifiers such as ROCKET and InceptionTime that extract features or learn deep representations, enabling you to train and evaluate models on temporal patterns in your data.

Feature extraction transforms raw time series into representations that capture temporal structure. Aeon includes methods like ROCKET, Catch22, and shapelet-based approaches that automatically generate features, reducing manual engineering and improving downstream classification or forecasting accuracy.

Yes. Aeon supports diverse time series tasks—forecasting, clustering, anomaly detection, and segmentation—through modular APIs (aeon.forecasting, aeon.clustering) on standardized data formats, letting you apply multiple analyses to sequential data without reformatting.

Aeon's deep learning architectures like InceptionTime work with standardized data shapes (n_cases, n_channels, n_timepoints) and accept transformed features from Catch22 or TSFresh. Preprocessing depends on your data; Aeon's modular design supports end-to-end pipelines with optional transformations.

Aeon includes distance-based methods and similarity search capabilities for time series, enabling you to find nearest neighbors, measure dissimilarity between sequences, and build clustering or retrieval workflows on sequential data like stock prices or wearable sensor streams.

Aeon's Python API supports pipelines that chain transformations (feature extraction), distance metrics, and classifiers or regressors. Use standardized data formats, modular components from aeon.classification and aeon.forecasting, and benchmarking tools to construct and evaluate complete workflows.