def _get_sequence_init(level, cluster_num, sequences1, sequences2, k, w):
distance = np.ones((cluster_num, len(sequences1)))
np.random.seed(RAND_SEED_INIT)
#init = np.array([4263, 3584]) #low
#init = np.array([6541, 1470]) #high
#init = np.append(init, np.random.randint(0, len(sequences1) - 1, 1))
init = np.random.randint(0, len(sequences1) - 1, 1)
print("--init 3:", init)
distance[0, :] = cskfuzzy.get_distance(level, w, sequences1, sequences2,
np.array(sequences1)[init],
np.array(sequences2)[init])
random.seed(RAND_SEED_INIT)
# get far away cluster initial
for c in range(1, cluster_num):
far_cluster = list(
np.where(
(distance[0:c, :] >= CLUSTER_DIST_THRESHOLD).all(axis=0))[0])
far_cluster = list(set(far_cluster) - set(init))
if len(far_cluster) == 0:
not_init = list(set(range(len(sequences1))) - set(init))
far_cluster = sorted(not_init,
key=lambda x: distance[0:c, x].sum(axis=0))
far_cluster = [far_cluster[-1]]
add_init = random.sample(far_cluster, 1)
distance[c, :] = cskfuzzy.get_distance(level, w, sequences1,
sequences2,
np.array(sequences1)[add_init],
np.array(sequences2)[add_init])
init = np.append(init, add_init)
print("[fuzzy c means]- get_init_u> \n-- init:", init)
# get enough k sequences for each cluster
#u = np.zeros((cluster_num, len(sequences1)))
filter_k = lambda row: row <= sorted(row)[k - 1]
large_k_indices = np.apply_along_axis(filter_k, axis=1, arr=distance)
u = large_k_indices.astype(int)
random.seed(RAND_SEED_K)
print("--each cluster initial # before random choose:", u.sum(axis=1))
for i in range(cluster_num):
if sum(u[i, :]) > k:
#print(np.where(u[i, :] == 1))
indices = set(np.where(u[i, :] == 1)[0])
indices.remove(init[i])
rand_k = random.sample(indices, k - 1)
u[i, :] = 0
u[i, :][rand_k] = 1
u[i, init[i]] = 1
print("--each cluster initial #:", u.sum(axis=1))
return u
def _distance_c_c(level, c, k, w, data1, data2, center):
center_data1 = np.array(data1)[center]
center_data2 = np.array(data2)[center]
distance = cskfuzzy.get_distance(level, w, center_data1, center_data2)
# distance = distance / np.amax(distance)
each_cluster = np.zeros((c, c))
np.fill_diagonal(each_cluster, 1)
each_cluster = np.repeat(each_cluster, k, axis=1)
d = each_cluster.dot(distance)
d = d.dot(each_cluster.transpose()) / (k * k)
np.fill_diagonal(d, 0)
return d
def _distance_c_c(level, c, k, w, data1, data2, center):
center_data1 = np.array(data1)[center]
center_data2 = np.array(data2)[center]
distance = cskfuzzy.get_distance(level, w, center_data1, center_data2)
#distance = distance / np.amax(distance)
each_cluster = np.zeros((c, c))
np.fill_diagonal(each_cluster, 1)
each_cluster = np.repeat(each_cluster, k, axis=1)
d = each_cluster.dot(distance)
d = d.dot(each_cluster.transpose()) / (k * k)
np.fill_diagonal(d, 0)
return d
