def dist_mean(data_set_1, data_set_2):
sum_dis = 0
for i in range(data_set_1.shape[0]):
for j in range(data_set_2.shape[0]):
sum_dis = sum_dis + hp.angDist(
(data_set_1[i, :], data_set_2[j, :]))
return sum_dis / (data_set_1.shape[0] * data_set_2.shape[0])
def dist_min(data_set_1, data_set_2):
min_dis = 1
for i in range(data_set_1.shape[0]):
for j in range(data_set_2.shape[0]):
dis_ij = hp.angDist(data_set_1[i, :], data_set_2[j, :])
if dis_ij < min_dis:
min_dis = dis_ij
return min_dis
def generateCluster(data_set, clus_idx):
clus_list = []
for i in range(data_set.shape[0]):
temp = []
for j in range(len(clus_idx)):
d = -1 * hp.angDist(data_set[i, :], data_set[clus_idx[j]])
temp.append(d)
c = clus_idx[temp.index(np.max(temp))]
clus_list.append(c)
return clus_list
def findRemotestIndex(data_mat, idx_list):
remotest_idx = 0
max_dist = 0
for i in range(data_mat.shape[0]):
min_dist = np.inf
for j in range(len(idx_list)):
if i == idx_list[j]:
min_dist = -np.inf
continue
cur_dist = hp.angDist(data_mat[i, :], data_mat[idx_list[j]])
if cur_dist < min_dist:
min_dist = cur_dist
if min_dist > max_dist:
max_dist = min_dist
remotest_idx = i
return remotest_idx
def angKmeans(data_set, k, create_cent=randCenter, calc_mean=hp.orientationMean, max_iterate=50,
min_error=0.0):
copy_set = data_set.copy()
m = copy_set.shape[0]
cluster_condition = np.zeros((m, 2))
centroids = create_cent(copy_set, k)
ini_centroids = centroids.copy()
cluster_changed = True
iterate_count = 0
old_error = 0
while cluster_changed:
iterate_count = iterate_count + 1
cluster_changed = False
for i in range(m):
min_dist = np.inf
min_index = -1
for j in range(k):
dist_ji = hp.angDist(centroids[j, :], copy_set[i, :])
if dist_ji < min_dist:
min_dist = dist_ji
min_index = j
if cluster_condition[i, 0] != min_index:
cluster_changed = True
cluster_condition[i, :] = min_index, min_dist ** 2
max_dist = -np.inf
max_index = -1
for i in range(m):
if cluster_condition[i, 1] > max_dist:
max_dist = cluster_condition[i, 1]
max_index = i
for cent in range(k):
pts_cluster = copy_set[np.nonzero(cluster_condition[:, 0] == cent)]
if pts_cluster.shape[0] == 0:
centroids[cent, :] = copy_set[max_index, :]
print("empty slice happened")
cluster_changed = True
else:
centroids[cent, :] = calc_mean(pts_cluster)
if iterate_count >= max_iterate:
cluster_changed = False
if np.abs(np.sum(cluster_condition[:, 1]) / m - old_error) <= min_error:
cluster_changed = False
old_error = np.sum(cluster_condition[:, 1]) / m
print("iterate round ", iterate_count, ", average error = ", np.sum(cluster_condition[:, 1]) / m)
return ini_centroids, centroids, cluster_condition
def calcSimilarity(data_set, p_mode=0):
similarity = np.mat(np.zeros((data_set.shape[0], data_set.shape[0])))
for i in range(data_set.shape[0]):
temp = np.mat(np.zeros((1, data_set.shape[0])))
for j in range(data_set.shape[0]):
s = -1 * hp.angDist(data_set[i, :], data_set[j, :])
temp[0, j] = s
similarity[i, :] = temp[0, :]
if p_mode == 0:
p = np.median(similarity.A)
elif p_mode == -1:
p = np.min(similarity)
elif p_mode == 1:
p = np.max(similarity)
else:
p = np.median(similarity)
for i in range(similarity.shape[0]):
similarity[i, i] = p
return similarity
def neighbour(data_set, data, eps):
nb_set = data_set.copy()
for i in range(data_set.shape[0] - 1, -1, -1):
if hp.angDist(data_set[i, :], data) > eps:
nb_set = np.delete(nb_set, i, axis=0)
return nb_set