Importance
polars_ts.importance
Permutation-based feature importance for time series. Closes #59.
permutation_importance(df, model, feature_cols, target_col='y', metric_fn=None, n_repeats=5, seed=42)
Compute permutation importance for each feature.
Shuffles each feature column and measures degradation in the scoring metric. Higher importance means the model relies more on that feature.
Parameters
df
Evaluation DataFrame with features and target.
model
A fitted sklearn-compatible estimator with predict.
feature_cols
Column names of the features.
target_col
Column with actual values.
metric_fn
Scoring function fn(df_with_y_and_y_hat) -> float.
Lower is better (e.g. MAE). Defaults to MAE.
n_repeats
Number of permutation repeats per feature.
seed
Random seed.
Returns
pl.DataFrame
DataFrame with columns ["feature", "importance_mean", "importance_std"],
sorted by importance (descending).