def test_threshold_adjacency_matrix(self):
adj_matrix = self.data_5d[0, 0, 0].squeeze()
mask, real_cost = util.threshold_adjacency_matrix(adj_matrix, 0)
npt.assert_equal(mask.sum(), 0)
npt.assert_equal(real_cost, 0)
mask, real_cost = util.threshold_adjacency_matrix(adj_matrix, 0.9)
npt.assert_equal(mask.sum(), 1800)
npt.assert_equal(real_cost, 0.9)
def test_threshold_adjacency_matrix(self):
adj_matrix = self.data_5d[0, 0, 0].squeeze()
mask, real_cost = util.threshold_adjacency_matrix(adj_matrix, 0)
npt.assert_equal(mask.sum(), 0)
npt.assert_equal(real_cost, 0)
mask, real_cost = util.threshold_adjacency_matrix(adj_matrix, .9)
npt.assert_equal(mask.sum(), 1800)
npt.assert_equal(real_cost, 0.9)
def test_find_true_cost(self):
adj_matrix = self.data_5d[0, 0, 0].squeeze()
mask, real_cost = util.threshold_adjacency_matrix(adj_matrix, 0.2)
true_cost = util.find_true_cost(mask)
npt.assert_equal(real_cost, true_cost)
## test on rounded array
adj_matrix = self.data_5d[0, 0, 0].squeeze().round(decimals=1)
mask, expected_cost = util.threshold_adjacency_matrix(adj_matrix, 0.2)
true_cost = util.find_true_cost(mask)
## the cost of the thresholded matrix will be less than expected
npt.assert_equal(real_cost > true_cost, True)
def test_find_true_cost(self):
adj_matrix = self.data_5d[0, 0, 0].squeeze()
mask, real_cost = util.threshold_adjacency_matrix(adj_matrix, 0.2)
true_cost = util.find_true_cost(mask)
npt.assert_equal(real_cost, true_cost)
## test on rounded array
adj_matrix = self.data_5d[0, 0, 0].squeeze().round(decimals=1)
mask, expected_cost = util.threshold_adjacency_matrix(adj_matrix, 0.2)
true_cost = util.find_true_cost(mask)
## the cost of the thresholded matrix will be less than expected
npt.assert_equal(real_cost > true_cost, True)
def calc_sa_modularity(mat, cost):
""" Use brainx to
generate binary matrix theshold based on cost
use simulated annealing to find communities
Parameters
----------
mat : numpy matrix
symmetric adjacency matrix, should not contain nan
cost: float
cost used to binarize adjacency matrix
Returns
-------
part : GraphPartition
part.index provides access to found communities
true_cost : float
the actual cost associated with thresholded adj matrix
"""
mask, real_cost = util.threshold_adjacency_matrix(mat, cost)
true_cost = util.find_true_cost(mask)
graph = nx.from_numpy_matrix(mask)
part, mod = modularity.simulated_annealing(graph)
return part, true_cost
def make_graph(matrix,cost=0.15):
mask, real_cost = util.threshold_adjacency_matrix(matrix, cost)
# true_cost = util.find_true_cost(mask)
graph = nx.from_numpy_matrix(mask)
return graph
