def _estimate_effect()

in dowhy/causal_estimators/econml.py [0:0]

• 58 lines of code
• 15 McCabe index (conditional complexity)

    def _estimate_effect(self):
        n_samples = self._treatment.shape[0]
        X = None  # Effect modifiers
        W = None  # common causes/ confounders
        Z = None  # Instruments
        Y = self._outcome
        T = self._treatment
        if self._effect_modifiers is not None:
            X = self._effect_modifiers
        if self._observed_common_causes_names:
            W = self._observed_common_causes
        if self.estimating_instrument_names:
            Z = self._estimating_instruments
        named_data_args = {'Y': Y, 'T': T, 'X': X, 'W': W, 'Z': Z}

        if self.estimator is None:
            estimator_class = self._get_econml_class_object(self._econml_methodname)
            self.estimator = estimator_class(**self.method_params["init_params"])
            # Calling the econml estimator's fit method
            estimator_argspec = inspect.getfullargspec(
                inspect.unwrap(self.estimator.fit))
            # As of v0.9, econml has some kewyord only arguments
            estimator_named_args = estimator_argspec.args + estimator_argspec.kwonlyargs
            estimator_data_args = {
                arg: named_data_args[arg] for arg in named_data_args.keys() if arg in estimator_named_args
                }
            if self.method_params["fit_params"] is not False:
                self.estimator.fit(**estimator_data_args,
                                   **self.method_params["fit_params"])

        X_test = X
        n_target_units = n_samples
        if X is not None:
            if type(self._target_units) is pd.DataFrame:
                X_test = self._target_units
            elif callable(self._target_units):
                filtered_rows = self._data.where(self._target_units)
                boolean_criterion = np.array(filtered_rows.notnull().iloc[:,0])
                X_test = X[boolean_criterion]
            n_target_units = X_test.shape[0]

        # Changing shape to a list for a singleton value
        if type(self._control_value) is not list:
            self._control_value = [self._control_value]
        if type(self._treatment_value) is not list:
            self._treatment_value = [self._treatment_value]
        T0_test = np.repeat([self._control_value], n_target_units, axis=0)
        T1_test = np.repeat([self._treatment_value], n_target_units, axis=0)
        est = self.estimator.effect(X_test, T0=T0_test, T1=T1_test)
        ate = np.mean(est)

        self.effect_intervals = None
        if self._confidence_intervals:
            self.effect_intervals = self.estimator.effect_interval(
                    X_test, T0=T0_test, T1=T1_test,
                    alpha=1-self.confidence_level)
        estimate = CausalEstimate(estimate=ate,
                                  control_value=self._control_value,
                                  treatment_value=self._treatment_value,
                                  target_estimand=self._target_estimand,
                                  realized_estimand_expr=self.symbolic_estimator,
                                  cate_estimates=est,
                                  effect_intervals=self.effect_intervals,
                                  _estimator_object=self.estimator)
        return estimate