def divide_score(class_1_data, class_1_label, class_2_data, class_2_label, K=None, score=mean_center_distance_score):
"""
use the above methods to evaluate the score of a certain partition
:param class_1_data:
:param class_1_label:
:param class_2_data:
:param class_2_label:
:param K:
:param score:
:return:
"""
if K is None:
K = len(np.unique(class_1_label)) + len(np.unique(class_2_label))
if 1 < len(np.unique(class_1_label)) < K - 1:
class_1_s = score(class_1_data, class_1_label)
else:
class_1_s = 0
if 1 < len(np.unique(class_2_label)) < K - 1:
class_2_s = score(class_2_data, class_2_label)
else:
class_2_s = 0
class_1_center = np.average(class_1_data, axis=0)
class_2_center = np.average(class_2_data, axis=0)
class_1_2_s = Distance.euclidean_distance(class_1_center, class_2_center)
if 1 < len(np.unique(class_1_label)) < K - 1:
confidence_score = class_1_2_s / (class_1_s + class_2_s)
else:
confidence_score = 0
return confidence_score
def divide_score(class_1_data, class_1_label, class_2_data, class_2_label, *, K=None, score=mean_center_distance_score):
"""
use the above methods to evaluate the score of a certain partition
:param class_1_data:
:param class_1_label:
:param class_2_data:
:param class_2_label:
:param K:
:param score:
:return:
"""
if K is None:
K = len(np.unique(class_1_label)) + len(np.unique(class_2_label))
if 1 < len(np.unique(class_1_label)) < K-1:
class_1_s = score(class_1_data, class_1_label)
else:
class_1_s = 0
if 1 < len(np.unique(class_2_label)) < K-1:
class_2_s = score(class_2_data, class_2_label)
else:
class_2_s = 0
class_1_center = np.average(class_1_data, axis=0)
class_2_center = np.average(class_2_data, axis=0)
class_1_2_s = Distance.euclidean_distance(class_1_center, class_2_center)
if 1 < len(np.unique(class_1_label)) < K-1:
confidence_score = class_1_2_s / (class_1_s + class_2_s)
else:
confidence_score = 0
return confidence_score
def mean_center_distance_score(class_data, class_label):
"""
calculate mean distance between different centers
:param class_data:
:param class_label:
:return:
"""
distance_sum = 0
class_center = []
for label in np.unique(class_label):
class_center.append(
np.average([
class_data[i]
for i in range(len(class_label)) if class_label[i] == label
],
axis=0))
for i in range(len(class_center)):
for j in range(i + 1, len(class_center)):
distance_sum = distance_sum + Distance.euclidean_distance(
class_center[i], class_center[j])
class_num = len(np.unique(class_label))
total_num = np.power(class_num, 2) - class_num
if total_num == 0:
total_num = 1
score = 2 * distance_sum / total_num
return score
def mean_distance_score(class_data, class_label):
"""
calculate mean distance between different samples
:param class_data:
:param class_label:
:return:
"""
center = np.mean(class_data, axis=0)
distance_sum = 0
for i in range(len(class_label)):
distance_sum = distance_sum + Distance.euclidean_distance(center, class_data[i])
return distance_sum / len(class_label)
def mean_distance_score(class_data, class_label):
"""
calculate mean distance between different samples
:param class_data:
:param class_label:
:return:
"""
center = np.mean(class_data, axis=0)
distance_sum = 0
for i in range(len(class_label)):
distance_sum = distance_sum + Distance.euclidean_distance(center, class_data[i])
return distance_sum / len(class_label)
def max_distance_score(class_data, class_label):
"""
calculate max distance between different samples
:param class_data:
:param class_label:
:return:
"""
max_distance = -np.inf
center = np.mean(class_data, axis=0)
for i in range(len(class_label)):
distance = Distance.euclidean_distance(center, class_data[i])
if distance > max_distance:
max_distance = distance
return max_distance
def min_distance_score(class_data, class_label):
"""
calculate min distance between different samples
:param class_data:
:param class_label:
:return:
"""
min_distance = np.inf
center = np.mean(class_data, axis=0)
for i in range(len(class_label)):
distance = Distance.euclidean_distance(center, class_data[i])
if distance < min_distance:
min_distance = distance
return min_distance
def agg_score(class_1_data, class_1_label, class_2_data, class_2_label, score=mean_distance_score):
"""
use the above methods to evaluate the score of a certain agglomeration
:param class_1_data:
:param class_1_label:
:param class_2_data:
:param class_2_label:
:param score:
:return:
"""
class_1_distance = score(class_1_data, class_1_label)
class_2_distance = score(class_2_data, class_2_label)
class_1_center = np.mean(class_1_data, axis=0)
class_2_center = np.mean(class_2_data, axis=0)
distance_between_two_class = Distance.euclidean_distance(class_1_center, class_2_center)
return 2 * distance_between_two_class / (class_1_distance + class_2_distance)
def agg_score(class_1_data, class_1_label, class_2_data, class_2_label, score=mean_distance_score):
"""
use the above methods to evaluate the score of a certain agglomeration
:param class_1_data:
:param class_1_label:
:param class_2_data:
:param class_2_label:
:param score:
:return:
"""
class_1_distance = score(class_1_data, class_1_label)
class_2_distance = score(class_2_data, class_2_label)
class_1_center = np.mean(class_1_data, axis=0)
class_2_center = np.mean(class_2_data, axis=0)
distance_between_two_class = Distance.euclidean_distance(class_1_center, class_2_center)
return 2 * distance_between_two_class / (class_1_distance + class_2_distance)
def min_center_distance_score(class_data, class_label):
"""
calculate min distance between different centers
:param class_data:
:param class_label:
:return:
"""
min_distance = np.inf
class_center = []
for label in np.unique(class_label):
class_center.append(np.average([class_data[i] for i in range(len(class_label)) if class_label[i] == label], axis=0))
for i in range(len(class_label)):
for j in range(i+1, len(class_center)):
distance = Distance.euclidean_distance(class_center[i], class_center[j])
if distance < min_distance:
min_distance = distance
return min_distance
def mean_center_distance_score(class_data, class_label):
"""
calculate mean distance between different centers
:param class_data:
:param class_label:
:return:
"""
distance_sum = 0
class_center = []
for label in np.unique(class_label):
class_center.append(np.average([class_data[i] for i in range(len(class_label)) if class_label[i] == label], axis=0))
for i in range(len(class_center)):
for j in range(i+1, len(class_center)):
distance_sum = distance_sum + Distance.euclidean_distance(class_center[i], class_center[j])
class_num = len(np.unique(class_label))
total_num = np.power(class_num, 2) - class_num
if total_num == 0:
total_num = 1
score = 2 * distance_sum / total_num
return score
