Clustering with KMeans and PCA

import numpy as np # linear algebra
import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)
import seaborn as sns
import matplotlib.pyplot as plt

import os
print(os.listdir("../input"))

df = pd.read_csv("../input/Wine.csv")
df.head()

# drop Customer_Segment
label = df.Customer_Segment
df = df.drop("Customer_Segment",axis = 1)
df.describe()

# checking NA
df.isnull().sum()

def plot_multi_variable(row,column) :
    """
    Plot multi variables histogram for dataframe with row = row and colunm = column
    """
    _,ax = plt.subplots(row,column,figsize = (12,14))
    for i in range(0,row) :
        for j in range(0,column) :
            if (row*i + j >= df.shape[1]) :
                break
            else :
                sns.distplot(df.iloc[:,row*i+j],color = "red",bins = 20,ax = ax[i,j])

plot_multi_variable(4,4)

def plot_boxplot_multi_variable(row,column) :
    """
    Plot multi variables boxplot for dataframe with row = row and colunm = column
    """
    _,ax = plt.subplots(row,column,figsize = (12,14))
    for i in range(0,row) :
        for j in range(0,column) :
            if (row*i + j >= df.shape[1]) :
                break
            else :
                sns.boxplot(y = df.iloc[:,row*i+j],color = "lightblue",ax = ax[i,j])

plot_boxplot_multi_variable(4,4)

_,_= plt.subplots(1,1,figsize = (10,10))
corr = df.corr()
mask = np.zeros_like(corr)
mask[np.triu_indices_from(mask)] = True
sns.heatmap(corr,mask = mask,cmap=plt.cm.PuOr,square = True,annot = True)

_,ax = plt.subplots(1,3,figsize = (18,3))
sns.regplot(df["Flavanoids"],df["Total_Phenols"],ax = ax[0],color = "red")
sns.regplot(df["Flavanoids"],df["OD280"],ax = ax[1],color = "green")
sns.regplot(df["Total_Phenols"],df["OD280"],ax = ax[2],color = "blue")

from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
from sklearn.cluster import KMeans

pl = Pipeline([
    ("scale",StandardScaler()),
    ("Kmeans", KMeans(n_clusters = 2,random_state = 13))
])

kmeans = pl.fit(df)

# centroids
print(kmeans.steps[1][1].cluster_centers_)
# withinss, Total within-cluster sum of squares , larger the better
print(kmeans.steps[1][1].inertia_)

# decide how many clusters is better
withinss = []
for i in range(1,10) :
    pl = Pipeline([
    ("scale",StandardScaler()),
    ("Kmeans", KMeans(n_clusters = i,random_state = 13))
    ])
    kmeans = pl.fit(df)
    withinss.append(kmeans.steps[1][1].inertia_)
plt.plot(withinss)
plt.ylabel("Total within-cluster sum of squares")
plt.xlabel("Number of Cluster")

pl_opt = Pipeline([
    ("scale",StandardScaler()),
    ("Kmeans", KMeans(n_clusters = 3,random_state = 13))
])

kmeans_opt = pl_opt.fit(df)
print(kmeans_opt.steps[1][1].inertia_)

df["Cluster"] = kmeans_opt.steps[1][1].labels_

# Centroids
centroids = kmeans_opt.steps[1][1].cluster_centers_
plt.scatter(centroids[:, 0], centroids[:, 1],
            marker='x', s=169, linewidths=3,
            color='r', zorder=10)

g = sns.PairGrid(df,hue = "Cluster",hue_kws={"marker": ["o", "s", "D"]},vars = df.columns[:df.shape[1]-1])
g = g.map_diag(sns.kdeplot)
g = g.map_upper(plt.scatter)
g = g.map_lower(sns.kdeplot)

from sklearn.decomposition import PCA
pipeline_pca = Pipeline([
    ("scale" , StandardScaler()),
    ("PCA" ,PCA(n_components = 13,random_state = 13))
])
pca_opt = pipeline_pca.fit_transform(df.drop("Cluster",axis = 1))

color = ["red" if i == 0 else "blue" if i == 1  else "yellow" for i in df["Cluster"]]
plt.figure(figsize = (7,7))
plt.scatter(pca_opt[:,0],pca_opt[:,2], c= color, alpha=0.8)
plt.title("PCA Decomposition with 3 Cluster")
plt.show()

from sklearn.metrics import classification_report
label_new = ["A" if i == 1 else "B" if i == 2 else "C" for i in label]
Cluster_new = ["A" if i == 2 else "B" if i == 0 else "C" for i in df["Cluster"]]
print(classification_report(Cluster_new,label_new))

# Accuracy
Correct = [1 if label_new[i] == Cluster_new[i] else 0 for i in range(0,len(label_new))]
print("Total Accuracy is :",np.mean(Correct))