2018-11-19-143147.ipynb

Kernel: Python 3 (Anaconda 5)

K-means

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import sklearn.cluster as skcl

attr_names = """sepal lenth
sepal width
petal length
petal width
class"""
attr_names = attr_names.split("\n")

iris_df = pd.read_csv("data/iris.data", sep=',', names=attr_names)

class_num = iris_df["class"]
class_names = class_num.unique()
class_names = list(class_names)
print(class_names)
class_num = class_num.apply(lambda val: class_names.index(val)+1)

['Iris-setosa', 'Iris-versicolor', 'Iris-virginica']

X1 = iris_df[attr_names[0]].values
X2 = iris_df[attr_names[1]].values
X3 = iris_df[attr_names[2]].values
X4 = iris_df[attr_names[3]].values
X = np.c_[X1,X2,X3,X4]
print(X.shape)

(150, 4)

Y = class_num.values
X23 = np.c_[X2, X3]

cl = skcl.KMeans(n_clusters=3)
cl.fit(X23)

KMeans(algorithm='auto', copy_x=True, init='k-means++', max_iter=300,
    n_clusters=3, n_init=10, n_jobs=None, precompute_distances='auto',
    random_state=None, tol=0.0001, verbose=0)

Z = cl.predict(X23)

plt.figure(figsize=(7,3))
plt.subplot(1,2,1)
plt.scatter(X2,X3, c=Y, edgecolors='k')
plt.subplot(1,2,2)
plt.scatter(X2,X3, c=Z, edgecolors='k')
centers = cl.cluster_centers_
plt.scatter(centers[:,0], centers[:,1], c='r', s=81)
plt.tight_layout()
plt.show()

import sklearn.decomposition as decomposition

pca = decomposition.PCA(n_components=2)
U = pca.fit_transform(X)

cl2 = skcl.KMeans(n_clusters=3)
cl2.fit(U)

KMeans(algorithm='auto', copy_x=True, init='k-means++', max_iter=300,
    n_clusters=3, n_init=10, n_jobs=None, precompute_distances='auto',
    random_state=None, tol=0.0001, verbose=0)

Z2 = cl2.predict(U)
U1, U2 = U[:,0], U[:,1]

plt.figure(figsize=(7,3))
plt.subplot(1,2,1)
plt.scatter(U1,U2, c=Y, edgecolors='k')
plt.subplot(1,2,2)
plt.scatter(U1,U2, c=Z2, edgecolors='k')
centers2 = cl2.cluster_centers_
plt.scatter(centers2[:,0], centers2[:,1], c='r', s=81)
plt.tight_layout()
plt.show()

cl3 = skcl.KMeans(n_clusters=2)
cl3.fit(U)

KMeans(algorithm='auto', copy_x=True, init='k-means++', max_iter=300,
    n_clusters=2, n_init=10, n_jobs=None, precompute_distances='auto',
    random_state=None, tol=0.0001, verbose=0)

Z3 = cl3.predict(U)
U1, U2 = U[:,0], U[:,1]

plt.figure(figsize=(8,4))
plt.subplot(1,2,1)
plt.scatter(U1,U2, c=Y, edgecolors='k')
plt.subplot(1,2,2)
plt.scatter(U1,U2, c=Z3, edgecolors='k')
centers3 = cl3.cluster_centers_
plt.scatter(centers3[:,0], centers3[:,1], c='r', s=81)
plt.tight_layout()
plt.show()

Задача

Разбить данные из wine dataset на кластеры (число кластеров совпадает с числом классов). Перед кластеризацией найти главные компоненты. Разбиение осуществлять относительно выбранных главных компонент.

attrs = ["class", "Alcohol", "Malic acid", "Ash", "Alcalinity of ash", "Magnesium", "Total phenols", "Flavanoids", "Nonflavanoid phenols", "Proanthocyanins", "Color intensity", "Hue", "OD280/OD315 of diluted wines", "Proline"]
wine_df = pd.read_csv("data/wine.data", sep=",", names = attrs)

class_num = wine_df["class"]
class_names = class_num.unique()

class_names = list(class_names)
print(class_names)
class_num = class_num.apply(lambda val: class_names.index(val)+1)

[1, 2, 3]

X1 = wine_df[attrs[1]].values
X2 = wine_df[attrs[2]].values
X3 = wine_df[attrs[3]].values
X4 = wine_df[attrs[4]].values
X5 = wine_df[attrs[5]].values
X6 = wine_df[attrs[6]].values
X7 = wine_df[attrs[7]].values
X8 = wine_df[attrs[8]].values
X9 = wine_df[attrs[9]].values
X10 = wine_df[attrs[10]].values
X11 = wine_df[attrs[11]].values
X12 = wine_df[attrs[12]].values
X13 = wine_df[attrs[13]].values

V1 = (X1-X1.min()) / (X1.max() - X1.min())
X1 = V1
V2 = (X2-X2.min()) / (X2.max() - X2.min())
X2 = V2
V3 = (X3-X3.min()) / (X3.max() - X3.min())
X3 = V3
V4 = (X4-X4.min()) / (X4.max() - X4.min())
X4 = V4
V5 = (X5-X5.min()) / (X5.max() - X5.min())
X5 = V5
V6 = (X6-X6.min()) / (X6.max() - X6.min())
X6 = V6
V7 = (X7-X7.min()) / (X7.max() - X7.min())
X7 = V7
V8 = (X8-X8.min()) / (X8.max() - X8.min())
X8 = V8
V9 = (X9-X9.min()) / (X9.max() - X9.min())
X9 = V9
V10 = (X10-X10.min()) / (X10.max() - X10.min())
X10 = V10
V11 = (X11-X11.min()) / (X11.max() - X11.min())
X11 = V11
V12 = (X12-X12.min()) / (X12.max() - X12.min())
X12 = V12
V13 = (X13-X13.min()) / (X13.max() - X13.min())
X13 = V13

X = np.c_[X1,X2,X3,X4,X5,X6,X7,X8,X9,X10,X11,X12,X13]
print(X.shape)

Y = class_num.values

pca = decomposition.PCA(n_components=13)
U = pca.fit_transform(X)

pca.explained_variance_ratio_

(178, 13)

array([0.40749485, 0.18970352, 0.08561671, 0.07426678, 0.05565301,
       0.04658837, 0.03663929, 0.02408789, 0.02274371, 0.02250965,
       0.01381292, 0.01273236, 0.00815095])

plt.figure(figsize=(5,3))
plt.bar(range(1,14), pca.explained_variance_ratio_)
plt.grid(1)
plt.minorticks_on()
plt.show()

pca = decomposition.PCA(n_components=2)
U = pca.fit_transform(X)

cl2 = skcl.KMeans(n_clusters=3)
cl2.fit(U)

KMeans(algorithm='auto', copy_x=True, init='k-means++', max_iter=300,
    n_clusters=3, n_init=10, n_jobs=None, precompute_distances='auto',
    random_state=None, tol=0.0001, verbose=0)

Z2 = cl2.predict(U)
U1, U2 = U[:,0], U[:,1]

plt.figure(figsize=(7,3))
plt.subplot(1,2,1)
plt.scatter(U1,U2, c=Y, edgecolors='k')
plt.subplot(1,2,2)
plt.scatter(U1,U2, c=Z2, edgecolors='k')
centers2 = cl2.cluster_centers_
plt.scatter(centers2[:,0], centers2[:,1], c='r', s=81)
plt.tight_layout()
plt.show()