def Laplace_Beltrami_Operator(file_name=default_VTK, which='norm1', kernel=cotangent_kernel, sigma=10, components=200, export=True, return_header=False):
""" Laplace_Beltrami_Operator."""
##############################################################
##### Extract Nodes and Meshes from VTK file:
Data = pyvtk.VtkData(file_name)
Nodes = np.asarray(Data.structure.points)
Meshes = np.asarray(Data.structure.polygons)
##############################################################
##### Compute Weight Matrix:
t0 = time()
aff_mat = compute_weights(Nodes, Meshes, kernel=kernel, add_to_graph=False, sigma=sigma)
print("done in %0.3fs" % (time() - t0))
##############################################################
##### Compute Graph Laplacian:
t0 = time()
Laplacian = graph_laplacian(aff_mat, which=which)
print("done in %0.3fs" % (time() - t0))
##############################################################
##### Obtain Eigen Spectrum:
print 'Finding the eigen spectrum...'
t0 = time()
Laplacian = Laplacian.tocsc()
if components >= Nodes.shape[0]:
components = Nodes.shape[0] - 1
evals, evecs = eigsh(Laplacian, components, sigma=0)
print("done in %0.3fs" % (time() - t0))
if export:
file_name2 = file_name[:-4] + '.npy'
h = open(file_name2, 'w')
pickle.dump(evals, h)
h.close()
################################################################
##### Write New VTK File With Feidler Vector Overlayed On Brain:
# f = vo.write_all(file_name[:-4]+'_eigen.vtk', Nodes, Meshes, evecs[:,1], label_type='Eigenvector')
#################################################################
##### Return Results:
if return_header:
return Nodes, Meshes, aff_mat, Laplacian, evals, evecs, Data.header
else:
return Nodes, Meshes, aff_mat, Laplacian, evals, evecs
def LabelPropagation(VTK = default_VTK, fundi_file = default_fundi_file, algorithm=default_algorithm, kernel=default_kernel, sigma=default_sigma, alpha=default_alpha, diagonal=0,
repeat=default_repeat, max_iters=3000, tol=1e-3, eps=1e-7):
"""
Main Function: Propagates labels to unlabeled nodes in graph.
Input: VTK file containing:
n x 3 Matrix X of 3-dimensional nodes.
n = l + u: l labeled nodes, u unlabeled nodes
m x 3 Matrix M of triangular meshes.
n x 1 Array of C unique labels.
C = number of classes
0 = no label
Output: Graph G with n labeled nodes along with probabilities.
Parameters
----------
kernel: function (x1, x2) --> float
choice of similarity metric
sigma: float
parameter for gaussian (rbf) kernel
alpha: float
clamping factor
diagonal: float
choice of diagonal entries in weight matrix
max_iters: float
maximum number of iterations allowed
tol: float
threshold to consider the system at steady state
eps: float
epsilon value for numerical stability in some algorithms.
"""
Nodes = Meshes = Labels = A = 0
# Step 0. Convert VTK file into numpy matrices
Data = pyvtk.VtkData(VTK)
Nodes = Data.structure.points
Meshes = Data.structure.polygons
data = convert_vtk_to_matrix(VTK, fundi_file)
# Step 1. Transform input data into graph G_basic
G_basic = build_graph(data['Nodes'])
# Step 2. Compute edge weights using connectivity established in VTK file
(G, aff_mat) = compute_weights(data['Nodes'], data['Meshes'], kernel, G=G_basic, sigma=sigma)
aff_mat = aff_mat.tocsr()
# Step 3. Transform column of labels into L x C Matrix, one column per class
(Label_Matrix, label_mapping) = get_label_matrix(data['Fundi_Labels'])
# Step 4. Propagate Labels!
if algorithm == "Weighted_Average":
print 'Performing Weighted Average Algorithm! Parameters: max_iters={0}'.format(str(max_iters))
(Graph, best_guess) = weighted_average(G, Label_Matrix, aff_mat, label_mapping, data['num_changed'],
repeat, diagonal, max_iters, tol)
elif algorithm == "Label_Spreading":
print 'Performing Label Spreading Algorithm! Parameters: alpha={0}, max_iters={1}'.format(str(alpha), str(max_iters))
Graph = label_spreading(G, Label_Matrix, aff_mat, label_mapping, data['num_changed'],
repeat, alpha, max_iters, tol)
elif algorithm == "Label_Propagation":
Graph = label_propagation(G, Label_Matrix, aff_mat, label_mapping, data['num_changed'],
repeat, alpha, eps, max_iters, tol)
elif algorithm == "Nearest_Neighbor":
print 'Performing 1_nearest_neighbor algorithm!'
Graph = nearest_neighbor(G, Label_Matrix, label_mapping)
else:
Graph = "That algorithm is not available."
# Step 5. Assess success of algorithm:
results = assess(Graph, data['Manual_Labels'], data['num_changed'], data['preserved_labels'], best_guess, label_mapping)
h = open('/home/eli/Desktop/NewLabels.vtk', 'w')
line_num = 0
total_lines_to_copy = 9 + data['Nodes'].shape[0] + data['Meshes'].shape[0]
for line in data['main_file']:
if line_num < total_lines_to_copy:
h.write(line)
line_num += 1
for i in xrange(len(Graph.nodes())):
h.write('{0}\n'.format(str(results[i])))
h.close()
return Graph, results
