Lbase3 = nx.linalg.spectrum.adjacency_spectrum(triFrame, weight='deg')
# reflect eigs about origin for laplacians
if laplacian:
xs = -xs
plt.hold(True)
#plt.plot(xs, np.abs(Lanalytic), 'g-', linewidth=2)
plt.plot(xs, np.abs(Lanalytic2), 'r-', linewidth=2)
plt.ylim((0, 0.5))
plt.xlim(xlim)
# A = tri.block_gaussian_guess(n, pA, pB)
# Lguess = eigvalsh(A, check_finite=False)
# Lguess /= np.sqrt(n)
# plt.hist(Lguess, bins=301, normed=True)
#from IPython import embed
if not laplacian:
P, K, Q = triFrame.base_matrices()
Keigs = eigvals(K)
Kmax = Keigs.max()
plt.plot((Kmax, Kmax), (0.01, 0.5), 'g--')
raw_input("Enter to continue")
#embed()
nx.write_gml(g, graph_file)
# from nonbacktracking import *
# B=nonbacktracking_matrix(g)
# from scipy.sparse.linalg import eigs as speigs
# B_eig1=speigs(B,k=1)[0][0]
# print B_eig1
# Bsm=nonbacktracking_matrix_guess(triFrame)
epsilon = 0.001
offset = 0.5
N = 400
num_eig = 10
d = 6
ps = {0: 0.5, 1: 0.5}
eset = [(0, 0), (0, 1), (1, 0), (1, 1)]
deg = {(0, 0): d + 9, (0, 1): d, (1, 0): d, (1, 1): d + 9}
g = Frame(nx.Graph(eset), p=ps, deg=deg)
gsamp = g.sample(N)
# rho = compute_spectrum(gsamp)
# rho.sort()
Pdiag, K, Q, P = g.base_matrices()
As = nx.adjacency_matrix(gsamp).todense()
I = nx.incidence_matrix(gsamp, oriented=True).todense()
D, V = eigh(K)
print "frame eigenvalues:"
print D
Ds, Vs = eigh(As)
Ds = np.flip(Ds, axis=0)
Vs = np.flip(Vs, axis=1)
print "ESD leading eigenvalues:"
print Ds[:num_eig]
Ps = As / np.tile(np.sum(As, axis=1), (1, As.shape[1])).astype(float)
DPs, VPs = eig(Ps)
Bs = non_backtracking_matrix(gsamp)
Bs = sp.csr_matrix(Bs)
