ML - Lab 7 - kmeans: metrics, elbow method, silhouette, heatmap

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import sklearn
import math


# Read Data - Split into target and features = cdata
cdata = pd.read_csv("./cdata.txt")
print("cdata summary = ")
print(cdata.describe())
print()
print()
target = cdata.loc[:, "Y"]
cdata = cdata.loc[:, ["X1", "X2"]]


# Draw the different classes
plt.figure()
plt.scatter(cdata[target == 1].X1, cdata[target == 1].X2, color="red", marker="o", label="1")
plt.scatter(cdata[target == 2].X1, cdata[target == 2].X2, color="blue", marker="o", label="2")
plt.scatter(cdata[target == 3].X1, cdata[target == 3].X2, color="green", marker="o", label="3")
plt.title("Initial Data")
plt.xlabel("X1")
plt.ylabel("X2")
plt.legend()
plt.show()


# Elbow method - Cohesion = SSE
from sklearn.cluster import KMeans
sse = []
RANGE = range(1, 11)
for i in RANGE:
    # sse list will contain the inertia_ of KMeans or cohesion or SSE
    sse.append(KMeans(n_clusters=i, init=cdata.loc[0:i-1, :]).fit(cdata).inertia_)

plt.figure()
plt.plot(RANGE, sse)                        # Solid line
plt.scatter(RANGE, sse, marker="o")         # Points
plt.title("SSE = Cohesion for every K")
plt.xlabel("K")
plt.ylabel("SSE")
plt.show()


# Select K = 3 for kmeans and calculate metrics
K = 3
kmeans = KMeans(n_clusters=K, init=cdata.loc[0:K-1, :])
kmeans = kmeans.fit(cdata)
cohesion = kmeans.inertia_
print("K = " + str(K))
print("Centroids = ")
print(kmeans.cluster_centers_)
print("Labels = " + str(kmeans.labels_))
print("Cohesion = " + str(cohesion))

separation = 0
distance = lambda x1, x2: math.sqrt(((x1.X1 - x2.X1) ** 2) + ((x1.X2 - x2.X2) ** 2))
m = cdata.mean()
for i in list(set(kmeans.labels_)):
    mi = cdata.loc[kmeans.labels_ == i, :].mean()
    Ci = len(cdata.loc[kmeans.labels_ == i, :].index)
    separation += Ci * (distance(m, mi) ** 2)
print("Separation = " + str(separation))
print("CSS + BSS = " + str(cohesion + separation))
print()
print()


# Draw Centroids and Points with the same color
plt.figure()
plt.scatter(cdata.X1, cdata.X2, c=kmeans.labels_)
plt.scatter(kmeans.cluster_centers_[:, 0], kmeans.cluster_centers_[:, 1], marker="+", s=169, c="black")
plt.title("Clustering with  K = " + str(K))
plt.xlabel("X1")
plt.ylabel("X2")
plt.show()


# Silhouette and Silhouette Plot
from sklearn.metrics import silhouette_samples, silhouette_score
print("Average Silhouette for each cluster: ")
for i in range(K):
    print(np.mean(silhouette_samples(cdata, kmeans.labels_)[kmeans.labels_ == i]))
print("Average Silhouette for the whole clustering = " + str(silhouette_score(cdata, kmeans.labels_)))
print()


# Silhouette Visualizer - Yellowbrick Library
import yellowbrick
from yellowbrick.cluster import SilhouetteVisualizer
visualizer = SilhouetteVisualizer(kmeans, colors='yellowbrick')
visualizer.fit(cdata)
visualizer.show()


# Heatmap
cdata["cluster"] = kmeans.labels_
cdata = cdata.sort_values("cluster").drop("cluster", axis=1)
from scipy.spatial import distance_matrix
dist = distance_matrix(cdata, cdata)
plt.imshow(dist, cmap='hot')
plt.colorbar()
plt.show()