def chi_square_distance(data, data_instance1, data_instance2):
"""This distance metric is used only for categorical data."""
if len(data_instance1.feature_vector) != len(data_instance2.feature_vector):
print("The two points must be defined in the same dimensional space.")
return
column_list = pick_up_all_data_columns(data)
total_attrs = []
for i, data_list in enumerate(column_list):
if i != 0:
total_attrs.append(sum(data_list))
average_profiles = [value/sum(total_attrs) for value in total_attrs]
relative_data = []
for row in data:
temp_sum = sum(row[1:]) #skips the label column
relative_data.append([feature/temp_sum for i,feature in enumerate(row) if i !=0])
column_list = pick_up_all_data_columns(relative_data)
cluster1_list = []
cluster2_list = []
for feature in data_instance1.feature_vector:
cluster1_list.append(feature/sum(data_instance1.feature_vector))
for feature in data_instance1.feature_vector:
cluster2_list.append(feature/sum(data_instance2.feature_vector))
sum_of_squares = 0.0
for i in range(0, len(cluster1_list)):
sum_of_squares += (1/average_profiles[i]) * pow(cluster1_list[i] - cluster2_list[i], 2)
return sqrt(sum_of_squares)
def standardised_euclidean_distance(data,data_instance1, data_instance2):
"""Returns the euclidean distance of two clusters in the n-dimensional space, which is defined by the number of their attributes,
weighted by the inverse of the corresponding attribute's variance.
"""
if len(data_instance1.feature_vector) != len(data_instance2.feature_vector):
print("The two points must be defined in the same dimensional space.")
return
column_list = pick_up_all_data_columns(data)
std_list = []
for i,column in enumerate(column_list):
if i != 0: #because the first column is always the data labels
std_list.append(standard_deviation(column))
sum_of_squares = 0.0
for i in range(0, len(data_instance1.feature_vector)):
sum_of_squares += 1/pow(std_list[i],2) * pow(data_instance1.feature_vector[i] - data_instance2.feature_vector[i], 2) #the weight here is the variance not the std
return sqrt(sum_of_squares)