QuantileSorter#

class openstef_models.transforms.postprocessing.quantile_sorter.QuantileSorter[source]

Bases: Transform[ForecastDataset, ForecastDataset]

Sort quantile forecasts to enforce monotonic ordering.

Probabilistic forecasts should have higher quantiles predict higher values (e.g., the 90th percentile should be greater than the 10th percentile). When quantiles are predicted independently, this property can be violated. This transform corrects violations by sorting quantile predictions for each time step.

The transform is stateless and requires no fitting.

Example

Basic usage with a forecast dataset

>>> import pandas as pd
>>> from datetime import timedelta
>>> from openstef_core.datasets import ForecastDataset
>>> # Create sample data with unsorted quantiles
>>> data = pd.DataFrame({
...     'quantile_P10': [1.0, 2.0, 3.0],
...     'quantile_P50': [0.5, 1.5, 2.5],  # Violates ordering
...     'quantile_P90': [2.0, 3.0, 4.0]
... }, index=pd.date_range('2025-01-01', periods=3, freq='h'))
>>> dataset = ForecastDataset(
...     data=data,
...     sample_interval=timedelta(hours=1)
... )
>>> sorter = QuantileSorter()
>>> sorted_dataset = sorter.transform(data=dataset)
>>> # Now quantile_P10 <= quantile_P50 <= quantile_P90 for each time step
>>> sorted_dataset.data.iloc[0].values.tolist()
[0.5, 1.0, 2.0]