Rocket classifier

polars_ts.classification.rocket_classifier

ROCKET and MiniROCKET classifiers for time series.

Uses existing ROCKET/MiniROCKET feature extraction + Ridge classifier for fast and accurate time series classification.

References

Dempster et al. (2020). ROCKET: Exceptionally fast and accurate time series classification using random convolutional kernels. DMKD.

RocketClassifier

Time series classifier using ROCKET features + Ridge.

Extracts ROCKET features from each series, then trains a Ridge classifier on the feature vectors.

Parameters

n_kernels Number of random convolutional kernels. alpha Ridge regularization strength. seed Random seed for kernel generation. id_col, target_col, time_col Column names.

fit(df, *, label_col='label')

Fit the classifier on training data.

Parameters

df Training DataFrame with id_col, target_col, and label_col. label_col Column containing class labels.

Returns

RocketClassifier Self, for chaining.

predict(df)

Predict class labels for test time series.

Parameters

df Test DataFrame with id_col and target_col.

Returns

pl.DataFrame Columns: unique_id, predicted_label.

MiniRocketClassifier

Time series classifier using MiniROCKET features + Ridge.

Same pattern as :class:RocketClassifier but uses the faster MiniROCKET feature extraction.

Parameters

n_kernels Number of kernels. alpha Ridge regularization strength. seed Random seed. id_col, target_col, time_col Column names.

fit(df, *, label_col='label')

Fit the classifier on training data.

predict(df)

Predict class labels for test time series.