Multiple lib changes

fluents/lib/nx_utils.py

@@ -201,16 +201,16 @@ def node_weighted_adj_matrix(G, weights=None, ave_type='harmonic', with_labels=F
 def weighted_adj_matrix(G, with_labels=False):
     """Adjacency matrix of an XGraph whos weights are given in edges.
     """
-    A,labels = NX.adj_matrix(G,with_labels=True)
+    A, labels = NX.adj_matrix(G, with_labels=True)
     W = A.astype('<f8')
-    for orf,i in labels.items():
-        for orf2,j in labels.items():
-            if G.has_edge(orf,orf2):
-                edge_weight = G.get_edge(orf,orf2)
-                W[i,j]=edge_weight
-                W[j,i]=edge_weight
+    for orf, i in labels.items():
+        for orf2, j in labels.items():
+            if G.has_edge(orf, orf2):
+                edge_weight = G.get_edge(orf, orf2)
+                W[i,j] = edge_weight
+                W[j,i] = edge_weight
     if with_labels==True:
-        return W,labels
+        return W, labels
     else:
         return W
 
@@ -418,6 +418,31 @@ def weighted_laplacian(G,with_labels=False):
     else:	
         return L            
 
+def subnetworks(G, T2):
+    """Return the highest scoring (T2-test) subgraph og G.
+
+    Use simulated annealing to identify highly scoring subgraphs.
+
+    ref: -- Ideker et.al (Bioinformatics 18, 2002)
+         -- Patil and Nielsen (PNAS 2006)
+    
+    """
+    N = 1000
+    states = [(node, False) for node in G.nodes()]
+    t2_last = 0.0
+    for i in xrange(N):
+        if i==0: #assign random states
+            states = [(state[0], True) for state in states if rand(1)>.5]
+        sub_nodes = [state[0] for state in states if state[1]]
+        Gsub = NX.subgraph(G, sub_nodes)
+        Gsub = NX.connected_components_subgraphs(Gsub)[0]
+        t2 = [T2[node] for node in Gsub]
+        if t2>t2_last:
+            pass
+        else:
+            p = numpy.exp()
+            
+    
 
 """Below are methods for calculating graph metrics
 
@@ -473,7 +498,7 @@ Ke = expm(A) .... expm(-A)?
 # 13.09.2206: update for use in numpy
 
 
-def K_expAdj(W, normalised=False, alpha=1.0):
+def K_expAdj(W, normalised=True, alpha=1.0):
     """Matrix exponential of adjacency matrix, mentioned in Kandola as a general diffusion kernel. 
     """
     W = asarray(W)
@@ -499,7 +524,7 @@ def K_expAdj(W, normalised=False, alpha=1.0):
     
     return dot(dot(vr,psigma),vri)
 
-def K_vonNeumann(W,normalised=False,alpha=1.0):
+def K_vonNeumann(W, normalised=True, alpha=1.0):
     """ The geometric series of path lengths.
     Returns matrix square root of pseudo inverse of the adjacency matrix.
     """
@@ -540,8 +565,8 @@ def K_laplacian(W, normalised=True, alpha=1.0):
     D = diag(sum(W,0))
     L = D - W
     if normalised==True:
-        T = diag(sqrt(1./sum(W,0)))
-        L = dot(dot(T,L),T)
+        T = diag(sqrt(1./sum(W, 0)))
+        L = dot(dot(T, L), T)
     e,vr = eig(L)
     e = real(e)
     vri = inv(vr)
@@ -551,7 +576,7 @@ def K_laplacian(W, normalised=True, alpha=1.0):
     for i in range(len(e)):
         if e[i] > cutoff:
             psigma[i] = 1.0/e[i]
-    K = dot(dot(vr,diag(psigma)),vri).astype(t)
+    K = dot(dot(vr, diag(psigma)), vri).astype(t)
     K = real(K)
     I = eye(n)
     K = (1-alpha)*I + alpha*K