remove ml_m from did_cs experimental

Author: SvenKlaassen

doubleml/did/did_cs.py

@@ -162,8 +162,13 @@ def trimming_threshold(self):
         return self._trimming_threshold
 
     def _initialize_ml_nuisance_params(self):
-        valid_learner = ['ml_g_d0_t0', 'ml_g_d0_t1',
-                         'ml_g_d1_t0', 'ml_g_d1_t1', 'ml_m']
+        if self.score == 'observational':
+            valid_learner = ['ml_g_d0_t0', 'ml_g_d0_t1',
+                             'ml_g_d1_t0', 'ml_g_d1_t1', 'ml_m']
+        else:
+            assert self.score == 'experimental'
+            valid_learner = ['ml_g_d0_t0', 'ml_g_d0_t1',
+                             'ml_g_d1_t0', 'ml_g_d1_t1']
         self._params = {learner: {key: [None] * self.n_rep for key in self._dml_data.d_cols}
                         for learner in valid_learner}
 

doubleml/did/tests/_utils_did_cs_manual.py

@@ -285,7 +285,6 @@ def fit_sensitivity_elements_did_cs(y, d, t, all_coef, predictions, score, in_sa
 
     for i_rep in range(n_rep):
 
-        m_hat = predictions['ml_m'][:, i_rep, 0]
         g_hat_d0_t0 = predictions['ml_g_d0_t0'][:, i_rep, 0]
         g_hat_d0_t1 = predictions['ml_g_d0_t1'][:, i_rep, 0]
         g_hat_d1_t0 = predictions['ml_g_d1_t0'][:, i_rep, 0]
@@ -305,6 +304,7 @@ def fit_sensitivity_elements_did_cs(y, d, t, all_coef, predictions, score, in_sa
         p_hat = np.mean(d)
         lambda_hat = np.mean(t)
         if score == 'observational':
+            m_hat = predictions['ml_m'][:, i_rep, 0]
             propensity_weight_d0 = np.divide(m_hat, 1.0-m_hat)
             if in_sample_normalization:
                 weight_d0t1 = np.multiply(d0t1, propensity_weight_d0)

doubleml/did/tests/test_did_cs_external_predictions.py

@@ -39,7 +39,8 @@ def doubleml_didcs_fixture(did_score, n_rep):
     ext_predictions["d"]["ml_g_d0_t1"] = dml_did_cs.predictions["ml_g_d0_t1"][:, :, 0]
     ext_predictions["d"]["ml_g_d1_t0"] = dml_did_cs.predictions["ml_g_d1_t0"][:, :, 0]
     ext_predictions["d"]["ml_g_d1_t1"] = dml_did_cs.predictions["ml_g_d1_t1"][:, :, 0]
-    ext_predictions["d"]["ml_m"] = dml_did_cs.predictions["ml_m"][:, :, 0]
+    if did_score == "observational":
+        ext_predictions["d"]["ml_m"] = dml_did_cs.predictions["ml_m"][:, :, 0]
 
     dml_did_cs_ext = DoubleMLDIDCS(ml_g=DMLDummyRegressor(), ml_m=DMLDummyClassifier(), **kwargs)
     dml_did_cs_ext.set_sample_splitting(all_smpls)