causalis.scenarios.unconfoundedness.refutation.score.score_validation

Score diagnostics focused on orthogonality and EIF stability.

Module Contents

Functions

run_score_diagnostics

Run orthogonality and influence diagnostics for ATE or ATTE scores.

Data

__all__

API

causalis.scenarios.unconfoundedness.refutation.score.score_validation.run_score_diagnostics(data: causalis.dgp.causaldata.CausalData, estimate: causalis.data_contracts.causal_estimate.CausalEstimate, *, overlap_threshold: Optional[float] = None, n_basis_funcs: Optional[int] = None, return_summary: bool = True) → Dict[str, Any]

Run orthogonality and influence diagnostics for ATE or ATTE scores.

The main object is the per-observation score contribution. For ATE, this diagnostic uses

.. math::

\hat\psi_i =
w_i(\hat g_1(X_i) - \hat g_0(X_i))
+ \bar w_i
\left[
(Y_i - \hat g_1(X_i)) \frac{D_i}{\hat m_i}
-
(Y_i - \hat g_0(X_i)) \frac{1-D_i}{1-\hat m_i}
\right]
- \hat\theta.

Good score behavior means:

• the empirical score average is close to zero,

• finite-basis derivatives with respect to nuisance parts are small,

• the influence distribution is not driven by a tiny number of very large :math:|\hat\psi_i|.

Parameters

data : CausalData Dataset used to fit the estimator. estimate : CausalEstimate Effect estimate with diagnostic_data containing nuisance predictions and optionally cached score arrays. overlap_threshold : float, optional Propensity clipping threshold. If omitted, the value is inferred from diagnostic or model metadata. n_basis_funcs : int, optional Number of simple basis functions used in orthogonality checks. Defaults to one intercept plus all available confounders. return_summary : bool, default True Include a compact summary table in the returned payload.

Returns

Dict[str, Any] Diagnostic report with orthogonality checks, influence summaries, optional out-of-sample tests, and a summary table.

Raises

ValueError If required diagnostic arrays are missing or have incompatible shapes.

Examples

from sklearn.ensemble import RandomForestClassifier, RandomForestRegressor from causalis.dgp import obs_linear_26_dataset from causalis.scenarios.unconfoundedness.model import IRM data = obs_linear_26_dataset( … n=1000, … seed=3141, … include_oracle=False, … return_causal_data=True, … ) irm = IRM( … data=data, … ml_g=RandomForestRegressor( … n_estimators=200, … max_depth=6, … min_samples_leaf=5, … random_state=3141, … ), … ml_m=RandomForestClassifier( … n_estimators=200, … max_depth=6, … min_samples_leaf=5, … random_state=3141, … ), … n_folds=3, … random_state=3141, … ) estimate = irm.fit().estimate(score=”ATE”) report = run_score_diagnostics(data, estimate) report[“summary”] # doctest: +SKIP report[“influence”][“top_influential”].head() # doctest: +SKIP

causalis.scenarios.unconfoundedness.refutation.score.score_validation.__all__

[‘run_score_diagnostics’]