clustering-of-csv-files/kMeansClustering.py

# conda activate PANDA

# https://machinelearningknowledge.ai/tutorial-for-k-means-clustering-in-python-sklearn/
# https://gist.githubusercontent.com/ryanorsinger/cb1222e506c1266b9cc808143ddbab82/raw/b2fe8213426159be7f9c8de108726d3d814153eb/mall_customers.csv

import os
nthreads = 1
os.environ["OMP_NUM_THREADS"] = str(nthreads)
'''
UserWarning: KMeans is known to have a memory leak on Windows with MKL,
when there are less chunks than available threads. You can avoid it by
setting the environment variable OMP_NUM_THREADS=1
'''
# https://stackoverflow.com/questions/73391779/setting-number-of-threads-in-python

'''
KMeans преимущества параллелизма на основе OpenMP через Cython. Небольшие порции данных (256 выборок)
обрабатываются параллельно, что, кроме того, снижает объем памяти. Дополнительные сведения о том,
как контролировать количество потоков
'''
# https://scikit-learn.ru/clustering/

from sklearn.cluster import KMeans
from sklearn import preprocessing
import sklearn.cluster as cluster
import sklearn.metrics as metrics
import pandas as pd
from sklearn.preprocessing import MinMaxScaler
import seaborn as sns
from matplotlib import pyplot as plt

df = pd.read_csv(r"Mall_Customers.csv")
print(df.head())
print('\n', df.shape)

scaler = MinMaxScaler()
scale = scaler.fit_transform(df[['annual_income','spending_score']])
df_scale = pd.DataFrame(scale, columns = ['annual_income','spending_score']);
print(df_scale.head(10))

km=KMeans(n_clusters=2)
# y_predicted = km.fit_predict(df[['annual_income','spending_score']])
y_predicted = km.fit_predict(df[['age','annual_income','spending_score']])
# y_predicted = km.fit_predict(df_scale[['annual_income','spending_score']])
print(y_predicted)


# центр тяжести двух кластеров:
print(km.cluster_centers_)

#print(km.predict(df[['annual_income','spending_score']]))

df['Clusters'] = km.labels_
print(df.head(21))

# sns.scatterplot(x="annual_income", y="spending_score", hue = 'Clusters',  data=df, palette='viridis')
# plt.scatter(km.cluster_centers_[0][0], km.cluster_centers_[0][1], label='центр тяжести кластерa 0')
# plt.scatter(km.cluster_centers_[1][0], km.cluster_centers_[1][1], label='центр тяжести кластерa 1', color = 'red')
# # plt.scatter(km.cluster_centers_[2][0], km.cluster_centers_[2][1], label='центр тяжести кластерa 2', color = 'violet')
# plt.grid()
# plt.legend()
# plt.show()

# # sns.scatterplot(x="age", y="spending_score", hue = 'Clusters',  data=df, palette='viridis')
# sns.scatterplot(x="age", y="annual_income", hue = 'Clusters',  data=df, palette='viridis')
# plt.scatter(km.cluster_centers_[0][0], km.cluster_centers_[0][1], label='центр тяжести кластерa 0')
# plt.scatter(km.cluster_centers_[1][0], km.cluster_centers_[1][1], label='центр тяжести кластерa 1', color = 'red')
# # plt.scatter(km.cluster_centers_[2][0], km.cluster_centers_[2][1], label='центр тяжести кластерa 2', color = 'violet')
# plt.grid()
# plt.legend()
# plt.show()



'''
Applying PCA
Now let us reduce the dimensionality of the dataset into two components.
'''

from sklearn.decomposition import PCA
pca = PCA(n_components=2)
# principalComponents = pca.fit_transform(df_scale)
principalComponents = pca.fit_transform(df[['age','annual_income','spending_score']])
pca_df = pd.DataFrame(data = principalComponents,
					columns = ['principal component 1', 'principal component 2'])
print(pca_df.head())

km=KMeans(n_clusters=2, init='k-means++')
y_predicted = km.fit_predict(pca_df)
print(y_predicted)

# центр тяжести двух кластеров:
print(km.cluster_centers_)

#print(km.predict(df[['annual_income','spending_score']]))

pca_df['Clusters'] = km.labels_
print(pca_df.head(21))

# sns.scatterplot(x="principal component 1", y="principal component 2", hue = 'Clusters',  data=pca_df, palette='viridis')
# plt.scatter(km.cluster_centers_[0][0], km.cluster_centers_[0][1], label='центр тяжести кластерa 0')
# plt.scatter(km.cluster_centers_[1][0], km.cluster_centers_[1][1], label='центр тяжести кластерa 1', color = 'red')
# # plt.scatter(km.cluster_centers_[2][0], km.cluster_centers_[2][1], label='центр тяжести кластерa 2', color = 'violet')
# plt.grid()
# plt.legend()
# plt.show()


'''
Finding Optimum Value of K
The Silhouette Method
Using the Silhouette method, it can be seen that the Silhouette value is maximum for K=6.
Hence it can be concluded that the dataset can be segmented properly with 7 clusters.
'''

for i in range(2,12):
    labels=cluster.KMeans(n_clusters=i,random_state=200,init='k-means++').fit(pca_df).labels_
    print ("Silhouette score for k(clusters) = "+str(i)+" is "
           +str(metrics.silhouette_score(pca_df,labels,metric="euclidean",sample_size=1000,random_state=200)))

'''
Silhouette score for k(clusters) = 2 is 0.3929352994462861
Silhouette score for k(clusters) = 3 is 0.45094150175950004
Silhouette score for k(clusters) = 4 is 0.49657104725411494
Silhouette score for k(clusters) = 5 is 0.4639067957511873
Silhouette score for k(clusters) = 6 is 0.5342952101268506
Silhouette score for k(clusters) = 7 is 0.45106742826242596
Silhouette score for k(clusters) = 8 is 0.4937480874760602
Silhouette score for k(clusters) = 9 is 0.4753608295862216
Silhouette score for k(clusters) = 10 is 0.4806759377217517
Silhouette score for k(clusters) = 11 is 0.46289709248542193
'''


kmeans = cluster.KMeans(n_clusters=5)
kmeans = kmeans.fit(pca_df)

pca_df['Clusters'] = kmeans.labels_

sns.scatterplot(x="principal component 1", y="principal component 2",hue = 'Clusters',  data=pca_df,palette='viridis')
plt.grid()
plt.show()