Fix SNMF/L to work with fixed initialization

Author: marinkaz

nimfa/methods/factorization/snmf.py

@@ -170,16 +170,14 @@ def factorize(self):
                 
         Return fitted factorization model.
         """
-        
-
         for run in range(self.n_run):
+            self.W, self.H = self.seed.initialize(self.V, self.rank, self.options)
             # in version SNMF/L, V is transposed while W and H are swapped and
             # transposed.
             if self.version == 'l':
                 self.V = self.V.T
-            
-            self.W, self.H = self.seed.initialize(
-                self.V, self.rank, self.options)
+                self.W, self.H = self.H.T, self.W.T
+
             if sp.isspmatrix(self.W):
                 self.W = self.W.tolil()
             if sp.isspmatrix(self.H):
@@ -222,7 +220,7 @@ def factorize(self):
                 self.W = self.W.tocsr()
             if sp.isspmatrix(self.H):
                 self.H = self.H.tocsr()
-            # transpose and swap the roles back if SNMF/L
+            # transpose and swap factors if SNMF/L
             if self.version == 'l':
                 self.V = self.V.T
                 self.W, self.H = self.H.T, self.W.T

nimfa/methods/seeding/fixed.py

@@ -49,7 +49,7 @@ def initialize(self, V, rank, options):
         if self.idx == 0:
             return self.W, self.H
         else:
-            self.W, getattr(self, 'H' + str(self.idx))
+            return self.W, getattr(self, 'H' + str(self.idx))
 
     def __repr__(self):
         return "fixed.Fixed()"