K-means clustering

K-means clustering.py

# Necessary packages are imported.
import numpy as np 
import pandas as pd 
import matplotlib.pyplot as plt
import seaborn as sns

# Loading of data set into 'cluster'.
cluster = pd.read_csv('k-means clustering.csv')

k = 3 # Number of clusters
rows = cluster.shape[0] # 'rows' contains the total number of rows in cluster data.
cols = cluster.shape[1] # 'cols' contains the total number of columns in cluster data.

centroids = cluster.loc[np.random.randint(1,rows+1,k)] # Randomly initialises 'k' no. of centroids.
centroids['new'] = list(range(1,k+1)) 
centroids.set_index('new',inplace = True) # New indices 1 to k are set for the dataframe 'centroids'.
d = np.random.rand(rows) # Initialization of d which would contain the centroid number closest to data point.

number_of_iterations = 5 
epsilon = list(range(number_of_iterations)) # 'epsilon' is the sum of squares of distances between points and centroid of a cluster for each iteration

for i in range(0,number_of_iterations): # This 'for' loop is for iterations.

    for j in range(0,rows): # This 'for' loop finds the centroid number closest to the data point.
        d[j] = ((centroids - cluster.loc[j])**2).sum(axis = 1).idxmin()
    cluster['centroid number'] = d # A new column 'centroid number' is added to dataframe 'cluster'.

    plt.subplots_adjust(bottom=0.1, right=2, top=0.9) # Adjusts the subplot.
    plt.subplot(1,number_of_iterations,i+1)
    sns.scatterplot(x = 'x',y = 'y',data = cluster,hue = 'centroid number',legend = 'full') # Scatter plot is plotted with differentiating factor as 'centroid number'
    plt.legend(bbox_to_anchor=(1, 1), loc=4, borderaxespad=0.5) #Adjusts the legend box.
    
    mean_x = list(range(k)) # Initialisation of 'mean_x' which will store mean of 'x' values of each cluster.
    mean_y = list(range(k)) # Initialisation of 'mean_y' which will store mean of 'y' values of each cluster.
    for m in range(0,k): # This 'for' loop calculates mean of 'x' and 'y' values of each cluster.
        mean_x[m] = cluster[cluster['centroid number'] == (m+1)]['x'].mean()
        mean_y[m] = cluster[cluster['centroid number'] == (m+1)]['y'].mean()
    centroids.replace(list(centroids['x']),mean_x,inplace = True) # The 'centroids' are replaced with the new values.
    centroids.replace(list(centroids['y']),mean_y,inplace = True)
    
    z = list(range(k)) # Initialisation of z  and centroid of each cluster.
    for p in range(0,k): # This 'for' loop calculates square of distances between data points and centroid of each cluster.
        z[p] = ((cluster[cluster['centroid number'] == p+1][['x','y']] - centroids.iloc[p])**2).values.sum()
    epsilon[i] = sum(z) # 'epsilon' is sum of squares of distances between points and centroid of a cluster for each iteration