metacausal.adapters.EconMLAdapter¶

class metacausal.adapters.EconMLAdapter(model, name=None, *, alpha=0.05, inference=None)[source]¶

Bases: object

Wrap an initialized-but-unfitted EconML estimator.

Supports any EconML estimator with .fit(Y, T, X=X) and .ate(X=X) or .effect(X=X) methods. The estimator template is deep-copied during fit() so that the original remains reusable for bootstrap resampling.

Parameters:

model – An initialized (unfitted) EconML estimator instance.
name (str | None) – Display name. Defaults to the class name.
alpha (float) – Significance level for analytical confidence intervals. Passed through to ate_interval() / effect_interval().
inference (Any) – Optional EconML fit-time inference backend, forwarded to model.fit(..., inference=inference).

Examples

>>> from econml.dml import LinearDML
>>> from metacausal.adapters import EconMLAdapter
>>> from metacausal.datasets import load_lalonde
>>> from sklearn.ensemble import (
...     HistGradientBoostingRegressor as HGBR,
...     HistGradientBoostingClassifier as HGBC,
... )
>>> X, T, Y = load_lalonde()
>>> dml = EconMLAdapter(
...     LinearDML(
...         model_y=HGBR(max_iter=20), model_t=HGBC(max_iter=20),
...         discrete_treatment=True, cv=2,
...     ),
... )
>>> dml.fit(X, T, Y, random_state=42)
>>> result = dml.ate()
>>> result
ComponentAteEstimate(ate=..., ci=[..., ...])

Methods

`__init__`
`ate`
`cate`
`fit`
`validate_outcome_type`	For binary outcomes, the wrapped EconML estimator must have `discrete_outcome=True` and a classifier in the slot that models the conditional outcome (`model_y` for `CausalForestDML`, `model_regression` for `DRLearner`).

Attributes

`name`
`supports_cate`

Details

ate(X=None)[source]¶

Parameters:: X (ndarray | None)
Return type:: ComponentAteEstimate

cate(X)[source]¶

Parameters:: X (ndarray)
Return type:: ComponentCateEstimate

fit(X, T, Y, **kwargs)[source]¶

Parameters:

Return type:

None

validate_outcome_type(detected)[source]¶

For binary outcomes, the wrapped EconML estimator must have discrete_outcome=True and a classifier in the slot that models the conditional outcome (model_y for CausalForestDML, model_regression for DRLearner).

Parameters:: detected (str)
Return type:: None

property name: str¶

property supports_cate: bool¶