fix linting and formular

Signed-off-by: kgao <[email protected]>

Author: kgao

econml/score/drscorer.py

@@ -10,14 +10,13 @@
 
 
 class DRScorer:
-    """ Scorer based on the DRLearner loss. Fits regression model g (using T-Learner) and propensity model p at fit time
-    and calculates the regression and propensity of the evaluation data::
+    """ Scorer based on the DRLearner loss. Fits regression model g (using T-Learner) and propensity model p at fit
+    time and calculates the regression and propensity of the evaluation data::
 
         g (model_regression) = E[Y | X, W, T]
-        
         p (model_propensity) = Pr[T | X, W]
 
-        Ydr(g,p) = g  + (Y - g ) / p * T
+        Ydr(g,p) = g(X,W,T) + (Y - g(X,W,T)) / p_T(X,W)
 
     Then for any given cate model calculates the loss::
 
@@ -206,17 +205,16 @@ def score(self, cate_model):
         score : double
             An analogue of the DR-square loss for the causal setting.
         """
-        g, p = self.drlearner_._cached_values.nuisances
-        Y =  self.drlearner_._cached_values.Y
-        T =  self.drlearner_._cached_values.T
-        Ydr = g  + (Y - g) / p * T
+        Y = self.drlearner_._cached_values.Y
+        T = self.drlearner_._cached_values.T
+        Y_pred, _ = self.drlearner_._cached_values.nuisances
+        Ydr = Y_pred[..., 1:] - Y_pred[..., [0]]
         X = self.drlearner_._cached_values.W[:, :self.dx_]
         sample_weight = self.drlearner_._cached_values.sample_weight
         if Ydr.ndim == 1:
             Ydr = Ydr.reshape((-1, 1))
 
         effects = cate_model.const_marginal_effect(X).reshape((-1, Ydr.shape[1]))
-        
         if sample_weight is not None:
             return 1 - np.mean(np.average((Ydr - effects)**2, weights=sample_weight, axis=0)) / self.base_score_
         else:

econml/tests/test_drscorer.py

@@ -29,9 +29,8 @@ def _get_data(self):
         X = np.random.normal(size=(1000, 3))
         T = np.random.binomial(2, scipy.special.expit(X[:, 0]))
         sigma = 0.001
-        y = (1 + .5*X[:, 0]) * T + X[:, 0] + np.random.normal(0, sigma, size=(1000,))
+        y = (1 + .5 * X[:, 0]) * T + X[:, 0] + np.random.normal(0, sigma, size=(1000,))
         return y, T, X, X[:, 0]
-    
 
     def test_comparison(self):
         def reg():
@@ -53,7 +52,7 @@ def clf():
                   ('dalearner', DomainAdaptationLearner(models=reg(), final_models=reg(), propensity_model=clf())),
                   ('slearner', SLearner(overall_model=reg())),
                   ('tlearner', TLearner(models=reg())),
-                  ('drlearner', DRLearner(model_propensity='auto',model_regression='auto',
+                  ('drlearner', DRLearner(model_propensity='auto', model_regression='auto',
                                           model_final=reg(), cv=3)),
                   ('rlearner', NonParamDML(model_y=reg(), model_t=clf(), model_final=reg(),
                                            discrete_treatment=True, cv=3)),
@@ -72,8 +71,8 @@ def clf():
                           multitask_model_final=False,
                           featurizer=None,
                           min_propensity=1e-6,
-                          cv=3, 
-                          mc_iters=2, 
+                          cv=3,
+                          mc_iters=2,
                           mc_agg='median')
         scorer.fit(Y_val, T_val, X=X_val)
         rscore = [scorer.score(mdl) for _, mdl in models]