Quantile regression

polars_ts.probabilistic.quantile_regression

Quantile regression forecaster for prediction intervals.

Trains one model per quantile level using any scikit-learn-compatible estimator that supports quantile loss (e.g. GradientBoostingRegressor with loss='quantile').

QuantileRegressor

Quantile regression forecaster producing prediction intervals.

Trains one model per quantile level. At prediction time, generates recursive multi-step forecasts for each quantile, using the median (q=0.5) prediction as the recursive input.

Parameters

estimator_factory Callable that takes a quantile level (float in (0, 1)) and returns a fresh scikit-learn-compatible estimator. Example: lambda q: GradientBoostingRegressor(loss='quantile', alpha=q). quantiles Quantile levels to predict (e.g. [0.1, 0.5, 0.9]). lags Lag offsets used as features (e.g. [1, 2, 7]). target_col Column with the target values. id_col Column identifying each time series. time_col Column with timestamps for ordering.

fit(df)

Fit one estimator per quantile on lag features derived from df.

All series are pooled together for a single global model per quantile.

Parameters

df Training DataFrame with at least id_col, time_col, and target_col.

Returns

QuantileRegressor Fitted regressor (self).

predict(df, h)

Generate h-step-ahead quantile forecasts by recursive prediction.

The median quantile (closest to 0.5) is used as the recursive input for subsequent steps.

Parameters

df DataFrame containing the history to predict from. h Forecast horizon (number of steps ahead).

Returns

pl.DataFrame DataFrame with columns [id_col, time_col, "y_hat", "q_0.1", ...]. The y_hat column uses the median quantile prediction. Quantile columns are named q_<level> for CRPS compatibility.