Bugfix: GMLVQ: Add missing transpose in computing final distance matrix

ceml/sklearn/lvq.py

@@ -62,14 +62,14 @@ def __init__(self, model, dist="l2"):
 
         # The model might have learned its own distance metric
         if isinstance(model, sklearn_lvq.GmlvqModel) or isinstance(model, sklearn_lvq.MrslvqModel):
-            self.dist_mat = create_tensor(np.dot(model.omega_, model.omega_))
+            self.dist_mat = create_tensor(np.dot(model.omega_.T, model.omega_))
         elif isinstance(model, sklearn_lvq.LgmlvqModel) or isinstance(model, sklearn_lvq.LmrslvqModel):
             if model.classwise == True:
-                self.omegas = [create_tensor(np.dot(omega, omega)) for omega in model.omegas_]
+                self.omegas = [create_tensor(np.dot(omega.T, omega)) for omega in model.omegas_]
                 self.classwise = True
                 self.dist_mats = None
             else:
-                self.dist_mats = [create_tensor(np.dot(omega, omega)) for omega in model.omegas_]
+                self.dist_mats = [create_tensor(np.dot(omega.T, omega)) for omega in model.omegas_]
                 self.classwise = False
 
         super(LVQ, self).__init__()

tests/sklearn/test_sklearn_lvq.py

@@ -20,8 +20,6 @@ def test_glvq():
 
     # Create and fit model
     model = GlvqModel(prototypes_per_class=3, max_iter=100, random_state=4242)
-    #model = GmlvqModel(prototypes_per_class=3, max_iter=200, random_state=4242)
-    #model = LgmlvqModel(prototypes_per_class=3, max_iter=100, random_state=4242)
     model.fit(X_train, y_train)
 
     # Select data point for explaining its prediction
@@ -73,8 +71,7 @@ def test_gmlvq():
     X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, random_state=4242)
 
     # Create and fit model
-    model = GmlvqModel(prototypes_per_class=3, max_iter=200, random_state=4242)
-    #model = LgmlvqModel(prototypes_per_class=3, max_iter=100, random_state=4242)
+    model = GmlvqModel(prototypes_per_class=3, max_iter=200, random_state=4242, dim=2)
     model.fit(X_train, y_train)
 
     # Select data point for explaining its prediction
@@ -84,7 +81,7 @@ def test_gmlvq():
     # Compute counterfactual
     features_whitelist = None
 
-    x_cf, y_cf, delta = generate_counterfactual(model, x_orig, 0, features_whitelist=features_whitelist, regularization="l1", C=1.0, optimizer="bfgs", return_as_dict=False)
+    x_cf, y_cf, delta = generate_counterfactual(model, x_orig, 0, features_whitelist=features_whitelist, regularization="l1", C=0.01, optimizer="bfgs", return_as_dict=False)
     assert y_cf == 0
     assert model.predict(np.array([x_cf])) == 0
 
@@ -102,7 +99,7 @@ def test_gmlvq():
 
 
     features_whitelist = [0, 1, 2, 3]
-    x_cf, y_cf, delta = generate_counterfactual(model, x_orig, 0, features_whitelist=features_whitelist, regularization="l1", C=1.0, optimizer="bfgs", return_as_dict=False)
+    x_cf, y_cf, delta = generate_counterfactual(model, x_orig, 0, features_whitelist=features_whitelist, regularization="l1", C=0.01, optimizer="bfgs", return_as_dict=False)
     assert y_cf == 0
     assert model.predict(np.array([x_cf])) == 0
     assert all([True if i in features_whitelist else delta[i] == 0. for i in range(x_orig.shape[0])])