Fixup: Basis and coefficient matrix have the same data type

README.rst

@@ -10,10 +10,11 @@ Documentation and examples on real-world data are at `Nimfa website`_.
 ****
 
 `Hidden patients and hidden genes - Understanding cancer data with matrix factorization`_ is
-a tutorial on how one can use Nimfa to analyze breast cancer transcriptome data sets from The
-International Cancer Genome Consortium (ICGC).
+a tutorial-like IPython notebook on how one can use Nimfa to analyze breast cancer transcriptome data sets from The
+International Cancer Genome Consortium (`ICGC`_).
 
 .. _Hidden patients and hidden genes - Understanding cancer data with matrix factorization: http://nbviewer.ipython.org/github/marinkaz/nimfa-ipynb/blob/master/ICGC%20and%20Nimfa.ipynb
+.. _ICGC: https://dcc.icgc.org
 
 Usage
 -----

nimfa/methods/factorization/nsnmf.py

@@ -157,11 +157,7 @@ def factorize(self):
         for run in range(self.n_run):
             self.W, self.H = self.seed.initialize(
                 self.V, self.rank, self.options)
-            self.S = sop(
-                (1 - self.theta) * sp.spdiags(
-                    [1 for _ in range(
-                        self.rank)], 0, self.rank, self.rank, 'csr'),
-                self.theta / self.rank, add)
+            self.S = np.diag(np.ones(self.rank) - self.theta) + self.theta / self.rank
             p_obj = c_obj = sys.float_info.max
             best_obj = c_obj if run == 0 else best_obj
             iter = 0
@@ -229,8 +225,8 @@ def update(self):
         H1 = repmat(W.sum(0).T, 1, self.V.shape[1])
         self.H = multiply(
             self.H, elop(dot(W.T, elop(self.V, dot(W, self.H), div)), H1, div))
-        # update basis matrix W
         H = dot(self.S, self.H)
+        # update basis matrix W
         W1 = repmat(H.sum(1).T, self.V.shape[0], 1)
         self.W = multiply(
             self.W, elop(dot(elop(self.V, dot(self.W, H), div), H.T), W1, div))

nimfa/methods/factorization/pmf.py

@@ -142,8 +142,7 @@ def factorize(self):
                 self.H, repmat(self.H.sum(axis=1), 1, self.V.shape[1]), div)
             self.v_factor = self.V.sum()
             self.V_n = sop(self.V.copy(), self.v_factor, div)
-            self.P = sp.spdiags(
-                [1. / self.rank for _ in range(self.rank)], 0, self.rank, self.rank, 'csr')
+            self.P = np.diag([1. / self.rank for _ in range(self.rank)])
             self.sqrt_P = sop(self.P, s=None, op=np.sqrt)
             p_obj = c_obj = sys.float_info.max
             best_obj = c_obj if run == 0 else best_obj

nimfa/models/nmf_ns.py

@@ -104,7 +104,7 @@ def distance(self, metric='euclidean', idx=None):
             return power(R, 2).sum()
         elif metric.lower() == 'kl':
             Va = dot(dot(self.W, self.S), self.H)
-            return (multiply(self.V, sop(elop(self.V, Va, div), op=log)) - self.V + Va).sum()
+            return (multiply(self.V, sop(elop(self.V, Va, div), op=np.log)) - self.V + Va).sum()
         else:
             raise utils.MFError("Unknown distance metric.")