def get_matrices(G):
'''
Parameters
----------
G : NetworkX graph
Returns
-------
T : Numpy array
L : NumPy array
Combinatorial Laplacian of G
W :
'''
W = nx.to_numpy_matrix(G, nodelist=sorted(G.nodes()))
L = nx.laplacian(G, nodelist=sorted(G.nodes()))
D = W.sum(1)
P = W / D
D = np.array(D).flatten()
Dsq = np.sqrt(D)
Disq = (1 / Dsq)
T = np.dot(np.diag(Dsq), P)
T = np.dot(T, np.diag(Disq))
T = (T + T.T) / 2 # Iron out numerical wrinkles
return T, L, np.array(W)
def MaxImpedanceComputation(InputGraph): MaxTotalImpedance=0 G=InputGraph.copy() number_of_vertices=G.order() vertexlist=G.nodes() for top_node in vertexlist: for ground_node in vertexlist: if ground_node<top_node: ordered_vertexlist=vertexlist[:] ordered_vertexlist.remove(top_node) ordered_vertexlist.remove(ground_node) ordered_vertexlist.insert(0,top_node) ordered_vertexlist.insert(0,ground_node) LaplacianMatrix=nx.laplacian(G,ordered_vertexlist) ConductanceMatrix=np.delete(LaplacianMatrix,0,0) ConductanceMatrix=np.delete(ConductanceMatrix,np.s_[0],1) InputVector=[0]*(number_of_vertices-1) InputVector[0]=1 VoltageVector=linalg.solve(ConductanceMatrix,InputVector) TotalImpedance=VoltageVector[0] if TotalImpedance>MaxTotalImpedance: MaxTotalImpedance=TotalImpedance return MaxTotalImpedance
def test_laplacian(self):
"Graph Laplacian"
NL=numpy.array([[ 3, -1, -1, -1, 0],
[-1, 2, -1, 0, 0],
[-1, -1, 2, 0, 0],
[-1, 0, 0, 1, 0],
[ 0, 0, 0, 0, 0]])
assert_equal(nx.laplacian(self.G),NL)
def findLaplacianKernel(g):
pm = nx.laplacian(g)
n = pm.shape[0]
#print "Number of nodes: ", n
e = np.matrix(np.ones(n)).transpose()
f = (e*e.transpose())/n
km = np.linalg.inv(pm - f) + f
return km
def _compute_C(G):
"""Compute inverse of Laplacian."""
try:
import numpy as np
except ImportError:
raise ImportError("flow_closeness_centrality() requires NumPy: http://scipy.org/ ")
L=nx.laplacian(G) # use ordering of G.nodes()
# remove first row and column
LR=L[1:,1:]
LRinv=np.linalg.inv(LR)
C=np.zeros(L.shape)
C[1:,1:]=LRinv
return C
def test1():
N = 3
G = nx.cycle_graph(N)
# Add self loop to all vertices
for node in G.nodes():
G.add_edge(node, node)
Ln_nx = nx.normalized_laplacian(G)
L_nx = nx.laplacian(G)
A = nx.adj_matrix(G)
# Using G.degree to compute D
D1 = np.array([G.degree()[n] for n in G])
Disqrt1 = np.array(1 / np.sqrt(D1))
Disqrt1 = np.diag(Disqrt1)
L1 = np.diag(D1) - A
Ln1 = np.dot(np.dot(Disqrt1, L1), Disqrt1)
# Using A.sum(1) to compute D
D2 = np.array(np.sum(A, 1)).flatten()
Disqrt2 = np.array(1 / np.sqrt(D2))
Disqrt2 = np.diag(Disqrt2)
L2 = np.diag(D2) - A
Ln2 = np.dot(np.dot(Disqrt2, L2), Disqrt2)
# Using G.degree and nx.laplacian to compute D
L3 = nx.laplacian(G)
Ln3 = np.dot(np.dot(Disqrt1, L3), Disqrt1)
Ln3_ = np.dot(Disqrt1, np.dot(L3, Disqrt1))
print "NetworkX version"
print Ln_nx
print "\nUsing G.degree to compute D"
print Ln1
print "\nUsing A.sum(1) to compute D"
print Ln2
print "\nUsing G.degree and nx.laplacian to compute D"
print Ln3
def _compute_C(G,weight='weight'):
"""Inverse of Laplacian."""
try:
import numpy as np
except ImportError:
raise ImportError(
"""current_flow_betweenness_centrality() requires NumPy
http://scipy.org/""")
L=nx.laplacian(G,weight=weight) # use ordering of G.nodes()
# remove first row and column
LR=L[1:,1:]
LRinv=np.linalg.inv(LR)
C=np.zeros(L.shape)
C[1:,1:]=LRinv
return C
def test_laplacian(self):
"Graph Laplacian"
NL = numpy.array([[3, -1, -1, -1, 0], [-1, 2, -1, 0, 0], [-1, -1, 2, 0, 0], [-1, 0, 0, 1, 0], [0, 0, 0, 0, 0]])
WL = 0.5 * NL
OL = 0.3 * NL
assert_equal(nx.laplacian(self.G), NL)
assert_equal(nx.laplacian(self.MG), NL)
assert_equal(nx.laplacian(self.G, nodelist=[0, 1]), numpy.array([[1, -1], [-1, 1]]))
assert_equal(nx.laplacian(self.WG), WL)
assert_equal(nx.laplacian(self.WG, weight=None), NL)
assert_equal(nx.laplacian(self.WG, weight="other"), OL)
def make_T(N, add_selfloops=False):
'''Compute the diffusion operator of an N-by-N grid graph.
The difussion operator is defined as
T = D^0.5 * P * D^-0.5, where D is the degree matrix, P=D^-1W is the random
walk matrix, and W is the adjancency matrix.
'''
G = nx.grid_2d_graph(N, N)
if add_selfloops:
for n in G:
G.add_edge(n, n)
T, _ = get_T(G)
L = nx.laplacian(G, nodelist=sorted(G.nodes()))
return T, L
def make_T(N, add_selfloops=False):
'''Compute the diffusion operator of an N-by-N grid graph.
The difussion operator is defined as
T = D^0.5 * P * D^-0.5, where D is the degree matrix, P=D^-1W is the random
walk matrix, and W is the adjancency matrix.
'''
G = nx.grid_2d_graph(N,N)
if add_selfloops:
for n in G:
G.add_edge(n, n)
T, _ = get_T(G)
L = nx.laplacian(G, nodelist=sorted(G.nodes()))
return T, L
def test_eigenvectors(self):
try:
import numpy as N
eigenval=N.linalg.eigvals
except ImportError:
raise SkipTest('NumPy not available.')
cs='ddiiddid'
G=nxt.threshold_graph(cs)
(tgeval,tgevec)=nxt.eigenvectors(cs)
dot=N.dot
assert_equal([ abs(dot(lv,lv)-1.0)<1e-9 for lv in tgevec ], [True]*8)
lapl=nx.laplacian(G)
tgev=[ dot(lv,dot(lapl,lv)) for lv in tgevec ]
assert_true(sum([abs(c-d) for c,d in zip(tgev,tgeval)]) < 1e-9)
tgev.sort()
lev=list(eigenval(lapl))
lev.sort()
assert_true(sum([abs(c-d) for c,d in zip(tgev,lev)]) < 1e-9)
def test_eigenvectors(self):
try:
import numpy as N
eigenval=N.linalg.eigvals
except ImportError:
raise SkipTest('NumPy not available.')
cs='ddiiddid'
G=nxt.threshold_graph(cs)
(tgeval,tgevec)=nxt.eigenvectors(cs)
dot=N.dot
assert_equal([ abs(dot(lv,lv)-1.0)<1e-9 for lv in tgevec ], [True]*8)
lapl=nx.laplacian(G)
tgev=[ dot(lv,dot(lapl,lv)) for lv in tgevec ]
assert_true(sum([abs(c-d) for c,d in zip(tgev,tgeval)]) < 1e-9)
tgev.sort()
lev=list(eigenval(lapl))
lev.sort()
assert_true(sum([abs(c-d) for c,d in zip(tgev,lev)]) < 1e-9)
def test_laplacian(self):
"Graph Laplacian"
NL = numpy.array([[3, -1, -1, -1, 0], [-1, 2, -1, 0, 0],
[-1, -1, 2, 0, 0], [-1, 0, 0, 1, 0], [0, 0, 0, 0,
0]])
WL = 0.5 * NL
OL = 0.3 * NL
assert_equal(nx.laplacian(self.G), NL)
assert_equal(nx.laplacian(self.MG), NL)
assert_equal(nx.laplacian(self.G, nodelist=[0, 1]),
numpy.array([[1, -1], [-1, 1]]))
assert_equal(nx.laplacian(self.WG), WL)
assert_equal(nx.laplacian(self.WG, weight=None), NL)
assert_equal(nx.laplacian(self.WG, weight='other'), OL)
def norm_lap(G, nodelist=None, weight="weight"):
# Graph or DiGraph, this is faster than above
if nodelist is None:
nodelist = G.nodes()
n = len(nodelist)
L = np.zeros((n, n))
deg = np.zeros((n, n))
index = dict((n, i) for i, n in enumerate(nodelist))
for ui, u in enumerate(nodelist):
totalwt = 0.0
for v, data in G[u].items():
try:
vi = index[v]
except KeyError:
continue
wt = data.get(weight, 1)
L[ui, vi] = -wt
if ui != vi:
totalwt += wt
print "tw", totalwt
L[ui, ui] = totalwt
if totalwt > 0.0:
deg[ui, ui] = np.sqrt(1.0 / totalwt)
Lnx = nx.laplacian(G)
print "Matriced used by NetworkX"
print "nx.laplacian(G)"
print Lnx
print "\nL"
print L
print "Disqrt"
print deg
L = np.dot(deg, np.dot(L, deg))
return L
def algebraic_distance(G, params={}):
'''
takes: graph G, computational parameters
returns:
a distance dictionary, d[node1][node2] giving the distance between the nodes
ref:
RELAXATION-BASED COARSENING AND
MULTISCALE GRAPH ORGANIZATION
DORIT RON, ILYA SAFRO, AND ACHI BRANDT
this code is not currently used in the main generator algorithm
wishlist: use sparse matrices
'''
H = nx.convert_node_labels_to_integers(G, discard_old_labels=False)
metric = params.get('metric', 'Linfinity')
num_relaxations_r = params.get('num_relaxations_r', 10)
num_test_vectors_K = params.get('num_test_vectors_K', 20)
lazy_walk_param_w = params.get('lazy_walk_param_w', 0.5)
if metric != 'Linfinity':
raise Exception('Metric other than Linifinity not implemented')
distance = {}
for node1 in H:
distance[node1] = {}
for node2 in H:
if node1 < node2: #save time
distance[node1][node2] = -np.inf
LAP = nx.laplacian(H)
diag_vec = np.diag(LAP)
DIAG = np.diag(diag_vec)
w_times_Dinv_times_D_minus_LAP = lazy_walk_param_w * np.dot(
np.diag([1. / el for el in diag_vec]), DIAG - LAP)
for t in xrange(num_test_vectors_K):
x = npr.rand(H.number_of_nodes(), 1)
for iteration in xrange(num_relaxations_r):
x = (1 - lazy_walk_param_w) * x + np.dot(
w_times_Dinv_times_D_minus_LAP, x)
for node1 in H:
for node2 in H:
dis = abs(x[node1] - x[node2])[0]
if node1 < node2 and dis > distance[node1][
node2]: #to save time, compute just the upper triangle of the matrix
distance[node1][node2] = dis
#generate the distance dictionary in the original node labels, and including the diagonal and lower triangle
ret = {}
for u in G:
ret[u] = {}
for v in G:
node1 = H.node_labels[u]
node2 = H.node_labels[v]
if node1 < node2:
ret[u][v] = distance[node1][node2]
elif node1 > node2:
ret[u][v] = distance[node2][node1]
else:
ret[u][v] = 0.
return ret
# Setup the node list
nodelist = [(i, j) for i in range(numx) for j in range(numy)]
def node_to_pos2d(node):
"""Give a node number, return the point on the physical mesh."""
return nodelist[node]
def pos_to_node2d(pos):
"""Give the position on the physical mesh, return the node."""
return nodelist.index(pos)
L = np.matrix(nx.laplacian(G, nodelist=nodelist))
# The rhs of the diffusion equation
rhs = -D * L / h**2
# The main temperature array.
T = np.matrix(np.zeros((nnodes, ntimes)))
# Setting initial temperature. Here we create a temporary array that has the
# physical shape. We then set the temp using x,y coordinates and then reshape
# for the computation. This rehaping matches the nodelist above.
Tnot = np.matrix(np.zeros((numx, numy)))
Tnot[2, 2] = 100.0
Tnot.shape = (nnodes, 1)
T[:, 0] = Tnot
import cvxopt as cvx
import cvxopt.lapack
import numpy as np
#make G undirected
G = nx.Graph(G)
#allocate weights to the edges
for (i, j) in G.edges():
G[i][j]['weight'] = c[i, j] + c[j, i]
maxcut = pic.Problem()
X = maxcut.add_variable('X', (N, N), 'symmetric')
#Laplacian of the graph
L = pic.new_param('L', 1 / 4. * nx.laplacian(G))
#ones on the diagonal
maxcut.add_constraint(pic.tools.diag_vect(X) == 1)
#X positive semidefinite
maxcut.add_constraint(X >> 0)
#objective
maxcut.set_objective('max', L | X)
#print maxcut
maxcut.solve(verbose=0)
#Cholesky factorization
V = X.value
for node in G.nodes():
G.add_edge(node, node)
Ln = norm_lap(G)
Ln_aw = norm_lap_aw(G)
print "\nNetworkX version"
print Ln
print "AJW version"
print Ln_aw
if __name__ == "__main__x":
N = 3
G = nx.cycle_graph(N)
print nx.laplacian(G)
print laplacian_matrix_aw(G)
nn
# Add self loop to all vertices
Gsl = G.copy()
for node in Gsl.nodes():
Gsl.add_edge(node, node)
print nx.laplacian(Gsl)
print laplacian_matrix_aw(Gsl)
if __name__ == "__main__":
test1()
test2()
import cvxopt.lapack
import numpy as np
#make G undirected
G=nx.Graph(G)
#allocate weights to the edges
for (i,j) in G.edges():
G[i][j]['weight']=c[i,j]+c[j,i]
maxcut = pic.Problem()
X=maxcut.add_variable('X',(N,N),'symmetric')
#Laplacian of the graph
L=pic.new_param('L',1/4.*nx.laplacian(G))
#ones on the diagonal
maxcut.add_constraint(pic.tools.diag_vect(X)==1)
#X positive semidefinite
maxcut.add_constraint(X>>0)
#objective
maxcut.set_objective('max',L|X)
#print maxcut
maxcut.solve(verbose = 0)
#Cholesky factorization
V=X.value
import networkx as nx
import numpy as np
import matplotlib.pyplot as plt
G = nx.Graph()
G.add_edges_from([(1, 2), (1, 3), (2, 3), (1, 4), (4, 5)])
G2 = nx.generators.random_graphs.fast_gnp_random_graph(10, .3)
# nx.adjacency_matrix(G)
L = nx.laplacian(G) # L=D-A
# np.linalg.eigvals(L)
np.linalg.eig(L)
res = nx.laplacian_spectrum(G)
print res
print nx.normalized_laplacian(G)
c = nx.communicability(G)
# drawing
nx.draw(G) # default using spring_layout: force-directed
# same as:
# pos = nx.spring_layout(G)
# nx.draw(G, pos)
nx.draw_random(G)
nx.draw_spectral(G)
plt.show()
plt.savefig('path.png')
plt.cla()
# random graph
G = nx.generators.random_graphs.random_regular_graph(6, 50)
plt.show()
def test_laplacian(self):
"Graph Laplacian"
NL = numpy.array([[3, -1, -1, -1, 0], [-1, 2, -1, 0, 0],
[-1, -1, 2, 0, 0], [-1, 0, 0, 1, 0], [0, 0, 0, 0,
0]])
assert_equal(nx.laplacian(self.G), NL)
import networkx as nx
import numpy as np
import matplotlib.pyplot as plt
G = nx.Graph()
G.add_edges_from([(1, 2), (1, 3), (2, 3), (1, 4), (4, 5)])
G2 = nx.generators.random_graphs.fast_gnp_random_graph(10, .3)
# nx.adjacency_matrix(G)
L = nx.laplacian(G) # L=D-A
# np.linalg.eigvals(L)
np.linalg.eig(L)
res = nx.laplacian_spectrum(G)
print res
print nx.normalized_laplacian(G)
c = nx.communicability(G)
# drawing
nx.draw(G) # default using spring_layout: force-directed
# same as:
# pos = nx.spring_layout(G)
# nx.draw(G, pos)
nx.draw_random(G)
nx.draw_spectral(G)
plt.show()
plt.savefig('path.png')
plt.cla()
# random graph
G = nx.generators.random_graphs.random_regular_graph(6, 50)
