def test_community_1():
"""
idea: regard a community as a sub-matrix, generate them respectively.
:return:
"""
node = 1000
lmd = 1.2
dmin = 1
dmax = 40
comu = 2
com_size = [int(node / comu) for _ in range(comu)]
com_size[-1] += node - sum(com_size)
i_list = [0] * node
j_list = [0] * node
total_edges = 0
img = np.zeros((node, node))
start_i, start_j = 0, 0
for size in com_size:
is_even = size % 2 == 1
power_law = NGPowerLaw(lmd, dmin, dmax, size)
for i in range(size):
d_out = power_law.get_d()
for _ in range(d_out):
j = power_law.get_j()
j = transform(is_even, j, size-1)
a_i = start_i + i
a_j = start_j + j
if img[a_i, a_j] == 0:
total_edges += 1
i_list[a_i] += 1
j_list[a_j] += 1
img[a_i, a_j] = 0.9
start_i += size
start_j += size
print('total edges: ', total_edges)
show_plot(i_list, math.log, dmin, dmax, title='out degree distribution')
show_plot(j_list, math.log, dmin, dmax, title='in degree distribution')
plt.imshow(img, cmap='gray')
plt.show()
def statistic_relation_data(self):
"""
show plot & show matrix thumbnail
:return:
"""
for rel in self.relation_ins:
if self.format == 'CSR':
row_name = 'row_offset.csr'
col_name = 'col_index.csr'
row_file = os.path.join(self.base_dir, rel.label, row_name)
col_file = os.path.join(self.base_dir, rel.label, col_name)
# show degree distribution
out_degree_list, in_degree_list = get_degree_list(
row_file, 'CSR', rel.node1, rel.node2, col_file)
if rel.out_distribution.get_d_type() == 'power_law':
show_plot(out_degree_list, math.log, rel.out_distribution.dmin, \
rel.out_distribution.dmax, 'out-degree distribution')
else:
show_plot(out_degree_list, lambda x: x, rel.out_distribution.dmin, \
rel.out_distribution.dmax, 'out-degree distribution')
if rel.in_distribution.get_d_type() == 'power_law':
show_plot(in_degree_list, math.log, rel.in_distribution.dmin, \
rel.in_distribution.dmax, 'in-degree distribution')
else:
show_plot(in_degree_list, lambda x: x, rel.in_distribution.dmin, \
rel.in_distribution.dmax, 'in-degree distribution')
# show matrix thumbnail
show_matrix_thumbnail(row_file, 'CSR', rel.node1, rel.node2,
col_file)
else:
data_name = 'data.' + self.format.lower()
data_file = os.path.join(self.base_dir, rel.label, data_name)
# show degree distribution
out_degree_list, in_degree_list = get_degree_list(
data_file, self.format, rel.node1, rel.node2)
if rel.out_distribution.get_d_type() == 'power_law':
show_plot(out_degree_list, math.log, rel.out_distribution.dmin, \
rel.out_distribution.dmax, 'out-degree distribution')
else:
show_plot(out_degree_list, lambda x: x, rel.out_distribution.dmin, \
rel.out_distribution.dmax, 'out-degree distribution')
if rel.in_distribution.get_d_type() == 'power_law':
show_plot(in_degree_list, math.log, rel.in_distribution.dmin, \
rel.in_distribution.dmax, 'in-degree distribution')
else:
show_plot(in_degree_list, lambda x: x, rel.in_distribution.dmin, \
rel.in_distribution.dmax, 'in-degree distribution')
# show matrix thumbnail
show_matrix_thumbnail(data_file, self.format, rel.node1,
rel.node2)
def test_overlap_community():
"""
idea: regard a community as a sub-matrix, generate them respectively.
when d_out > threshold * dmax, add some noise
the degree of noise: [1, dmax]
x: degree
f(x): the probability of degree x
f(x) = a * exp(-x / c)
procedure: generate noise degree
y = random(0, 1)
d = -c * ln(exp(-1/c) - y * ( (exp(-1/c) - exp(-dmax/c) ) )
parameter description:
c: > 0, the bigger c, the more noise
:return:
"""
node = 1000
lmd = 1.2
dmin = 1
dmax = 40
comu = 3
threshold = 0.5
overlap = 0.1
param_c = 10
overlap_size = int(node * overlap / 2)
com_size = [int(node / comu) for _ in range(comu)]
com_size[-1] += node % comu
i_list = [0] * node
j_list = [0] * node
total_edges = 0
extra_edges = 0
expect_edges = 0
img = np.zeros((node, node))
start_i, start_j = 0, 0
threshold = round(threshold * dmax)
const_a = math.exp(-1/param_c)
const_b = const_a - math.exp(-dmax/param_c)
overlap_pl = NGPowerLaw(lmd*2, 1, int((dmin+dmax)/4), overlap_size)
ol_even = overlap_size % 2 == 1
for size in com_size:
is_even = size % 2 == 1
power_law = NGPowerLaw(lmd, dmin, dmax, size)
for i in range(size):
d_out = power_law.get_d()
expect_edges += d_out
a_i = start_i + i
for _ in range(d_out):
j = power_law.get_j()
j = transform(is_even, j, size - 1)
a_j = start_j + j
if img[a_i, a_j] == 0:
total_edges += 1
i_list[a_i] += 1
j_list[a_j] += 1
img[a_i, a_j] = 0.9
if d_out > threshold:
y = random.random()
d_extra = int(-param_c * math.log(const_a - y * const_b))
extra_edges += d_extra
for _ in range(d_extra):
j = power_law.get_j()
j = transform(is_even, j, size - 1)
j = scale(j, 0, size - 1, 0, node - size - 1)
if j > start_j:
j = j + size
if img[a_i, j] == 0:
total_edges += 1
i_list[a_i] += 1
j_list[j] += 1
img[a_i, j] = 0.9
if start_j > overlap_size and overlap_size < size:
o_start_i = start_i
o_start_j = start_j - overlap_size
for _i in range(overlap_size):
d_over = overlap_pl.get_d()
expect_edges += d_over
a_i = _i + o_start_i
for _ in range(d_over):
j = overlap_pl.get_j()
j = transform(ol_even, j, overlap_size-1)
a_j = j + o_start_j
if img[a_i, a_j] == 0:
total_edges += 1
img[a_i, a_j] = 0.9
i_list[a_i] += 1
j_list[a_j] += 1
if start_i > overlap_size and overlap_size < size:
o_start_i = start_i - overlap_size
o_start_j = start_j
for _i in range(overlap_size):
d_over = overlap_pl.get_d()
expect_edges += d_over
a_i = _i + o_start_i
for _ in range(d_over):
j = overlap_pl.get_j()
j = transform(ol_even, j, overlap_size-1)
a_j = j + o_start_j
if img[a_i, a_j] == 0:
total_edges += 1
img[a_i, a_j] = 0.9
i_list[a_i] += 1
j_list[a_j] += 1
start_i += size
start_j += size
print('total edges: ', total_edges)
print('extra edges: ', extra_edges)
print('expect edges: ', expect_edges)
show_plot(i_list, math.log, dmin, dmax, title='out degree distribution')
show_plot(j_list, math.log, dmin, dmax, title='in degree distribution')
plt.imshow(img, cmap='gray')
plt.show()
def test_add_noise():
"""
idea: regard a community as a sub-matrix, generate them respectively.
add noise: when d_out > threshold, add some noise
:return:
"""
node = 1000
lmd = 1.2
dmin = 1
dmax = 40
comu = 3
threshold = 0.5
com_size = [int(node / comu) for _ in range(comu)]
com_size[-1] += node % comu
i_list = [0] * node
j_list = [0] * node
total_edges = 0
extra_edges = 0
expect_edges = 0
img = np.zeros((node, node))
start_i, start_j = 0, 0
threshold *= dmax
e1 = math.exp(1)
ed = math.exp(dmax)
for size in com_size:
is_even = size % 2 == 1
power_law = NGPowerLaw(lmd, dmin, dmax, size)
for i in range(size):
d_out = power_law.get_d()
expect_edges += d_out
a_i = start_i + i
for _ in range(d_out):
j = power_law.get_j()
j = transform(is_even, j, size-1)
a_j = start_j + j
if img[a_i, a_j] == 0:
total_edges += 1
i_list[a_i] += 1
j_list[a_j] += 1
img[a_i, a_j] = 0.9
if d_out > threshold:
y = random.random()
d_extra = int(-math.log((1-y)/e1 + y/ed))
extra_edges += d_extra
for _ in range(d_extra):
j = power_law.get_j()
j = transform(is_even, j, size-1)
j = scale(j, 0, size-1, 0, node-size-1)
if j > start_j:
j = j + size
if img[a_i, j] == 0:
total_edges += 1
i_list[a_i] += 1
j_list[j] += 1
img[a_i, j] = 0.9
start_i += size
start_j += size
print('total edges: ', total_edges)
print('extra edges: ', extra_edges)
print('expect edges: ', expect_edges)
show_plot(i_list, math.log, dmin, dmax, title='out degree distribution')
show_plot(j_list, math.log, dmin, dmax, title='in degree distribution')
plt.imshow(img, cmap='gray')
plt.show()
def test_community():
node = 1000
lmd = 1.23
dmin = 1
dmax = 20
power_law = NGPowerLaw(lmd, dmin, dmax, node)
# two communities
d_list = [0] * node
i_list = [0] * node
j_list = [0] * node
for i in range(node):
d_list[i] = power_law.get_d()
img = np.zeros((node, node))
is_even = node % 2 == 1
step0 = int(node / 2)
size = int(node / 2)
for i in range(step0):
if d_list[i] >= size:
extra = d_list[i] - size
for j in range(size):
img[i, j] = 0.5
j_list[j] += 1
i_list[i] += size
for _ in range(extra):
j = power_law.get_j()
j = transform(is_even, j, node-1)
j = scale(j, 0, node-1, size, node-1)
if img[i, j] == 0:
i_list[i] += 1
img[i, j] = 0.5
j_list[j] += 1
else:
for _ in range(d_list[i]):
j = power_law.get_j()
j = transform(is_even, j, node-1)
j = scale(j, 0, node-1, 0, size-1)
if img[i, j] == 0:
i_list[i] += 1
img[i, j] = 0.5
j_list[j] += 1
for i in range(step0, node):
if d_list[i] >= size:
extra = d_list[i] - size
for j in range(step0, node):
img[i, j] = 0.5
j_list[j] += 1
i_list[i] += size
for _ in range(extra):
j = power_law.get_j()
j = transform(is_even, j, node-1)
j = scale(j, 0, node-1, 0, size-1)
if img[i, j] == 0:
i_list[i] += 1
img[i, j] = 0.5
j_list[j] += 1
else:
for _ in range(d_list[i]):
j = power_law.get_j()
j = transform(is_even, j, node-1)
j = scale(j, 0, node-1, step0, node-1)
if img[i, j] == 0:
i_list[i] += 1
img[i, j] = 0.5
j_list[j] += 1
show_plot(i_list, math.log, dmin, dmax, title='out degree distribution')
show_plot(j_list, math.log, dmin, dmax, title='in degree distribution')
plt.imshow(img, cmap='gray')
plt.show()
points.append((6, (x, y, z + .15)))
#
x = float(sys.argv[1])
y = float(sys.argv[2])
z = float(sys.argv[3])
# example 0, 0, 0.438
# later on, need to input initial angles, or the first 6 inputs, plus the 3 inputs of the coordinates to get to
#
def handler(signum, frame):
raise Exception("end of time")
# for i in range(0, 20):
# for j in range(0, 20):
# for k in range(0, 20):
# signal.signal(signal.SIGALRM, handler)
# signal.alarm(20)
# try:
# angles = run_km(i/10, j/10, k/10)
# print(i/10, j/10, k/10, "OK")
# except Exception as exc:
# print(i/10, j/10, k/10, "timed out")
for i in range(1, 5):
angles = run_km(x/4*i, y/4*i, z/4*i)
print(110+i)
visualization.show(math.radians(angles[1]), math.radians(angles[0]), math.radians(angles[3]), math.radians(angles[2]), x, y, z)
visualization.show_plot()
def statistic(self, filename, n, fmt):
out_degree_list, in_degree_list = get_degree_list(filename, fmt, n, n)
show_plot(out_degree_list, lambda x: x, self.min_outd, self.max_outd, 'out-degree distribution')
show_plot(in_degree_list, lambda x: x, self.min_ind, self.max_ind, 'in-degree distribution')