4.1.8. doubleml.utils.PSProcessor#

class doubleml.utils.PSProcessor(clipping_threshold: float = 0.01, extreme_threshold: float = 1e-12, calibration_method: str | None = None, cv_calibration: bool = False)#

Processor for propensity score calibration, clipping, and validation.

Parameters:

clipping_threshold (float, default=1e-2) – Minimum and maximum bound for propensity scores after clipping.
extreme_threshold (float, default=1e-12) – Threshold below which propensity scores are considered extreme. Used for generating warnings.
calibration_method ({'isotonic', None}, optional) – If provided, applies the specified calibration method to the propensity scores before clipping.
cv_calibration (bool, default=False) – Whether to use cross-validation for calibration. Only applies if a calibration method is specified.

Examples

>>> import numpy as np
>>> from doubleml.utils import PSProcessor
>>> ps = np.array([0.001, 0.2, 0.5, 0.8, 0.999])
>>> treatment = np.array([0, 1, 1, 0, 1])
>>> processor = PSProcessor(clipping_threshold=0.01)
>>> adjusted = processor.adjust_ps(ps, treatment)
>>> print(np.round(adjusted, 3))
[0.01 0.2  0.5  0.8  0.99]

Methods

`adjust_ps`(propensity_scores, treatment[, ...])	Adjust propensity scores via calibration and clipping.
`from_config`(config)	Create PSProcessor from PSProcessorConfig.

Attributes

`calibration_method`	Get the calibration method.
`clipping_threshold`	Get the clipping threshold.
`cv_calibration`	Get whether cross-validation calibration is used.
`extreme_threshold`	Get the extreme threshold.

PSProcessor.adjust_ps(propensity_scores: ndarray, treatment: ndarray, cv: int | list | None = None, learner_name: str | None = None) → ndarray#

Adjust propensity scores via calibration and clipping.

Parameters:

propensity_scores (np.ndarray) – Raw propensity score predictions.
treatment (np.ndarray) – Treatment assignments (1 for treated, 0 for control).
cv (int or list, optional) – Cross-validation strategy for calibration. Used only if calibration is applied.
learner_name (str, optional) – Name of the learner providing the propensity scores, used in warnings.

Returns:

Clipped and validated propensity scores.

Return type:

np.ndarray

classmethod PSProcessor.from_config(config: PSProcessorConfig)#: Create PSProcessor from PSProcessorConfig.