Merge pull request #27 from cdt15/develop

Develop

lingam/__init__.py

@@ -18,4 +18,4 @@
            'CausalEffect', 'VARLiNGAM', 'VARMALiNGAM', 'LongitudinalLiNGAM', 'LongitudinalBootstrapResult',
            'BottomUpParceLiNGAM', 'RCD']
 
-__version__ = '1.5.2'
+__version__ = '1.5.3'

lingam/base.py

@@ -133,20 +133,22 @@ def _estimate_adjacency_matrix(self, X, prior_knowledge=None):
         self : object
             Returns the instance itself.
         """
-        sink_vars = get_sink_variables(prior_knowledge)
-        exo_vars = get_exo_variables(prior_knowledge)
+        if prior_knowledge is not None:
+            pk = prior_knowledge.copy()
+            np.fill_diagonal(pk, 0)
 
         B = np.zeros([X.shape[1], X.shape[1]], dtype='float64')
         for i in range(1, len(self._causal_order)):
             target = self._causal_order[i]
             predictors = self._causal_order[:i]
 
-            # target is not used for prediction if it is included in exogenous variables
-            if target in exo_vars:
-                continue
+            # Exclude variables specified in no_path with prior knowledge
+            if prior_knowledge is not None:
+                predictors = [p for p in predictors if pk[target, p] != 0]
 
-            # sink variables are not used as predictors
-            predictors = [v for v in predictors if v not in sink_vars]
+            # target is exogenous variables if predictors are empty
+            if len(predictors) == 0:
+                continue
 
             B[target, predictors] = predict_adaptive_lasso(
                 X, predictors, target)

lingam/bottom_up_parce_lingam.py

@@ -143,6 +143,12 @@ def _extract_partial_orders(self, pk):
             no_path_pairs = pairs[counts < 2]
 
         check_pairs = np.concatenate([path_pairs, no_path_pairs[:, [1, 0]]])
+        if len(check_pairs) == 0:
+            # If no pairs are extracted from the specified prior knowledge, 
+            # discard the prior knowledge.
+            self._Aknw = None
+            return None
+
         pairs = np.unique(check_pairs, axis=0)
         return pairs[:, [1, 0]]  # [to, from] -> [from, to]
 
@@ -286,22 +292,24 @@ def _estimate_adjacency_matrix(self, X, prior_knowledge=None):
         self : object
             Returns the instance itself.
         """
-        sink_vars = get_sink_variables(prior_knowledge)
-        exo_vars = get_exo_variables(prior_knowledge)
+        if prior_knowledge is not None:
+            pk = prior_knowledge.copy()
+            np.fill_diagonal(pk, 0)
 
         B = np.zeros([X.shape[1], X.shape[1]], dtype='float64')
         for i in range(1, len(self._causal_order)):
             target = self._causal_order[i]
 
-            # target is not used for prediction if it is included in exogenous variables
-            if target in exo_vars:
-                continue
-
             # Flatten the array into one dimension
             predictors = self._flatten(self._causal_order[:i])
 
-            # sink variables are not used as predictors
-            predictors = [v for v in predictors if v not in sink_vars]
+            # Exclude variables specified in no_path with prior knowledge
+            if prior_knowledge is not None:
+                predictors = [p for p in predictors if pk[target, p] != 0]
+
+            # target is exogenous variables if predictors are empty
+            if len(predictors) == 0:
+                continue
 
             B[target, predictors] = predict_adaptive_lasso(
                 X, predictors, target)

lingam/direct_lingam.py

@@ -123,6 +123,12 @@ def _extract_partial_orders(self, pk):
             no_path_pairs = pairs[counts < 2]
 
         check_pairs = np.concatenate([path_pairs, no_path_pairs[:, [1, 0]]])
+        if len(check_pairs) == 0:
+            # If no pairs are extracted from the specified prior knowledge, 
+            # discard the prior knowledge.
+            self._Aknw = None
+            return None
+
         pairs = np.unique(check_pairs, axis=0)
         return pairs[:, [1, 0]]  # [to, from] -> [from, to]
 

lingam/var_lingam.py

@@ -76,7 +76,7 @@ def fit(self, X):
         lingam_model = self._lingam_model
         if lingam_model is None:
             lingam_model = DirectLiNGAM()
-        elif not issubclass(lingam_model, _BaseLiNGAM):
+        elif not isinstance(lingam_model, _BaseLiNGAM):
             raise ValueError('lingam_model must be a subclass of _BaseLiNGAM')
 
         M_taus = self._ar_coefs

lingam/varma_lingam.py

@@ -84,7 +84,7 @@ def fit(self, X):
         lingam_model = self._lingam_model
         if lingam_model is None:
             lingam_model = DirectLiNGAM()
-        elif not issubclass(lingam_model, _BaseLiNGAM):
+        elif not isinstance(lingam_model, _BaseLiNGAM):
             raise ValueError('lingam_model must be a subclass of _BaseLiNGAM')
 
         phis = self._ar_coefs

tests/test_direct_lingam.py

@@ -182,7 +182,7 @@ def test_prior_knowledge_invalid():
 
     # prior knowledge: invalid
     pk = np.array([
-        [0, -1, -1],
+        [0, -1, 1],
         [-1,  0, -1],
         [-1, -1,  0],
         [-1, -1, -1],

tests/test_multi_group_direct_lingam.py

@@ -275,7 +275,7 @@ def test_prior_knowledge_invalid():
 
     # prior knowledge: invalid
     pk = np.array([
-        [0, -1, -1],
+        [0, -1, 1],
         [-1,  0, -1],
         [-1, -1,  0],
         [-1, -1, -1],