metacausal.aggregation.NuisanceEstimates

class metacausal.aggregation.NuisanceEstimates(e_hat, mu1_hat, mu0_hat)[source]

Bases: object

Out-of-fold nuisance model predictions.

Variables:

  • e_hat (array of shape (n,)) – Propensity scores P(T=1|X), trimmed to [propensity_trim, 1-propensity_trim]. Already trimmed at fit time — do not untrim before computing pseudo-outcomes.

  • mu1_hat (array of shape (n,)) – Out-of-fold predictions of E[Y|X, T=1].

  • mu0_hat (array of shape (n,)) – Out-of-fold predictions of E[Y|X, T=0].

Parameters:

Notes

For TrainAvgSplit, positions corresponding to the training set are np.nan. Only the averaging-set positions (fold_spec.test_indices[0]) are populated.

Methods

__init__

Attributes

m_hat

Marginal conditional mean E[Y|X] = e*mu1 + (1-e)*mu0.

e_hat

mu1_hat

mu0_hat

Details

e_hat: ndarray
property m_hat: ndarray

Marginal conditional mean E[Y|X] = e*mu1 + (1-e)*mu0.

mu0_hat: ndarray
mu1_hat: ndarray