import numpy as np

import pandas as pd

from sklearn.preprocessing import StandardScaler

from sklearn.metrics import davies_bouldin_score
from scipy.spatial.distance import cdist

import pyclustering.cluster.center_initializer as pyci

pip install pyclustering

Requirement already satisfied: pyclustering in c:\users\manvi\appdata\

local\programs\python\python311\lib\site-packages (0.10.1.2)
Requirement already satisfied: scipy>=1.1.0 in c:\users\manvi\appdata\
local\programs\python\python311\lib\site-packages (from pyclustering)
(1.11.3)
Requirement already satisfied: matplotlib>=3.0.0 in c:\users\manvi\
appdata\local\programs\python\python311\lib\site-packages (from
pyclustering) (3.8.2)
Requirement already satisfied: numpy>=1.15.2 in c:\users\manvi\
appdata\local\programs\python\python311\lib\site-packages (from
pyclustering) (1.26.4)
Requirement already satisfied: Pillow>=5.2.0 in c:\users\manvi\
appdata\local\programs\python\python311\lib\site-packages (from
pyclustering) (9.5.0)
Requirement already satisfied: contourpy>=1.0.1 in c:\users\manvi\
appdata\local\programs\python\python311\lib\site-packages (from
matplotlib>=3.0.0->pyclustering) (1.2.0)
Requirement already satisfied: cycler>=0.10 in c:\users\manvi\appdata\
local\programs\python\python311\lib\site-packages (from
matplotlib>=3.0.0->pyclustering) (0.12.1)
Requirement already satisfied: fonttools>=4.22.0 in c:\users\manvi\
appdata\local\programs\python\python311\lib\site-packages (from
matplotlib>=3.0.0->pyclustering) (4.47.2)
Requirement already satisfied: kiwisolver>=1.3.1 in c:\users\manvi\
appdata\local\programs\python\python311\lib\site-packages (from
matplotlib>=3.0.0->pyclustering) (1.4.5)
Requirement already satisfied: packaging>=20.0 in c:\users\manvi\
appdata\local\programs\python\python311\lib\site-packages (from
matplotlib>=3.0.0->pyclustering) (23.2)
Requirement already satisfied: pyparsing>=2.3.1 in c:\users\manvi\
appdata\local\programs\python\python311\lib\site-packages (from
matplotlib>=3.0.0->pyclustering) (3.1.1)
Requirement already satisfied: python-dateutil>=2.7 in c:\users\manvi\
appdata\local\programs\python\python311\lib\site-packages (from
matplotlib>=3.0.0->pyclustering) (2.8.2)
Requirement already satisfied: six>=1.5 in c:\users\manvi\appdata\
local\programs\python\python311\lib\site-packages (from python-
dateutil>=2.7->matplotlib>=3.0.0->pyclustering) (1.16.0)
Note: you may need to restart the kernel to use updated packages.
data = pd.read_csv("Credit Card Customer Data.csv")
data.head()

Sl_No Customer Key Avg_Credit_Limit Total_Credit_Cards \

0 1 87073 100000 2
1 2 38414 50000 3
2 3 17341 50000 7
3 4 40496 30000 5
4 5 47437 100000 6

Total_visits_bank Total_visits_online Total_calls_made

0 1 1 0
1 0 10 9
2 1 3 4
3 1 1 4
4 0 12 3

import numpy as np
import pandas as pd
from sklearn.preprocessing import StandardScaler
from sklearn.metrics import davies_bouldin_score
from scipy.spatial.distance import cdist

class PROCLUS:
def __init__(self, k, l, max_iter=10):
"""
PROCLUS Subspace Clustering Algorithm
:param k: Number of clusters
:param l: Number of dimensions per cluster
:param max_iter: Maximum iterations for convergence
"""
self.k = k
self.l = l
self.max_iter = max_iter
self.medoids = None
self.subspaces = None
self.clusters = None

def initialize_medoids(self, data):

""" Select initial medoids using a greedy selection. """
np.random.seed(42)
indices = np.random.choice(len(data), self.k, replace=False)
self.medoids = data[indices]

def compute_subspaces(self, data, clusters):

""" Determine relevant dimensions for each cluster. """
subspaces = []
for cluster in clusters:
if len(cluster) > 0:
cluster_data = data[cluster]
 mean_dist = np.mean(np.abs(cluster_data -
np.mean(cluster_data, axis=0)), axis=0)
top_dims = np.argsort(mean_dist)[:self.l] # Select
lowest-variance dimensions
subspaces.append(top_dims)
else:
subspaces.append(np.random.choice(data.shape[1],
self.l, replace=False))
return subspaces

def assign_points(self, data, first_iter=False):

clusters = [[] for _ in range(len(self.medoids))]
for i, point in enumerate(data):
min_dist = float('inf')
best_cluster = -1
for cluster_id, medoid in enumerate(self.medoids):
if first_iter:
dist = np.linalg.norm(point - medoid) # full-
space distance
else:
subspace_dims = self.subspaces[cluster_id]
dist = np.sum(np.abs(point[subspace_dims] -
medoid[subspace_dims])) # Manhattan distance
if dist < min_dist:
min_dist = dist
best_cluster = cluster_id
clusters[best_cluster].append(i)
return clusters

def update_medoids(self, data, clusters):

""" Update medoids by selecting the most central point in each
cluster. """
new_medoids = []
for cluster in clusters:
if len(cluster) > 0:
cluster_data = data[cluster]
distances = np.sum(cdist(cluster_data, cluster_data,
metric='cityblock'), axis=1)
new_medoids.append(cluster_data[np.argmin(distances)])
else:
new_medoids.append(np.random.choice(data, 1)[0])
return np.array(new_medoids)

def fit(self, data):

self.initialize_medoids(data)

# Initial assignment before computing subspaces

initial_clusters = self.assign_points(data, first_iter=True)
 self.subspaces = self.compute_subspaces(data, initial_clusters)

for _ in range(self.max_iter):
clusters = self.assign_points(data)
self.subspaces = self.compute_subspaces(data, clusters)
new_medoids = self.update_medoids(data, clusters)

if np.all(new_medoids == self.medoids):
break
self.medoids = new_medoids

return clusters

# Dunn Index Calculation

def dunn_index(self,data, clusters):
intra_dists = []
inter_dists = []

# Ensure clusters are not empty

valid_clusters = [cluster for cluster in clusters if
len(cluster) > 1]

for cluster in valid_clusters:

cluster_data = data[cluster]
intra_dists.append(np.max(cdist(cluster_data,
cluster_data, metric='euclidean')))

for i in range(len(valid_clusters)):
for j in range(i + 1, len(valid_clusters)):

inter_dists.append(np.min(cdist(data[valid_clusters[i]],
data[valid_clusters[j]], metric='euclidean')))

if len(inter_dists) == 0 or len(intra_dists) == 0:
return float("inf") # Return infinity if calculation is
not possible

return min(inter_dists) / max(intra_dists)

# Load Sample Credit Card Data

df = pd.read_csv("Credit Card Customer Data.csv")
df.head()

Sl_No Customer Key Avg_Credit_Limit Total_Credit_Cards \

0 1 87073 100000 2
1 2 38414 50000 3
2 3 17341 50000 7
3 4 40496 30000 5
4 5 47437 100000 6

Total_visits_bank Total_visits_online Total_calls_made

0 1 1 0
1 0 10 9
2 1 3 4
3 1 1 4
4 0 12 3

scaler = StandardScaler()
data_scaled = scaler.fit_transform(df)
print(data_scaled)

[[-1.72942847 1.24691971 1.74018685 ... -0.86045063 -0.54748969

-1.25153737]
[-1.72417983 -0.65320273 0.41029254 ... -1.47373077 2.5205186
1.89185881]
[-1.71893118 -1.47609839 0.41029254 ... -0.86045063 0.13428993
0.1455276 ]
...
[ 1.71893118 -0.05105538 2.93709172 ... -0.86045063 2.17962879
-0.90227113]
[ 1.72417983 0.99629832 3.65523464 ... -0.86045063 4.22496765
-1.25153737]
[ 1.72942847 0.97657819 3.52224521 ... -1.47373077 3.20229822
-0.55300488]]

# Run PROCLUS
k = 5 # Number of clusters
l = 3 # Subspace dimensionality
proclus = PROCLUS(k, l)
clusters = proclus.fit(data_scaled)

# Assign Labels
labels = np.zeros(len(data_scaled))
for cluster_id, cluster in enumerate(clusters):
for index in cluster:
labels[index] = cluster_id

# Compute Cluster Quality

db_index = davies_bouldin_score(data_scaled, labels)
dunn = proclus.dunn_index(data_scaled, clusters)

print(f"Davies-Bouldin Index: {db_index}")

print(f"Dunn Index: {dunn}")
Davies-Bouldin Index: 2.1216181090876955
Dunn Index: 0.05855920277922305