吴恩达机器学习python作业之K-means

参考链接：吴恩达|机器学习作业7.0.k-means聚类_学吧学吧终成学霸的博客-CSDN博客
任务一：实现K-means算法

任务二：使用K-means算法进行图像压缩

任务一：实现K-means算法

numpy中的norm()函数求范数_若水cjj的博客-CSDN博客_np.norm

import numpy as np
from matplotlib import pyplot as plt
from scipy import io
import random
#1.读取数据
dt = io.loadmat("E:\机器学习\吴恩达\data_sets\ex7data2.mat")
x = dt["X"] #(300,2)

#2.可视化数据集
"""plt.scatter(x[:,0],x[:,1],marker="+")
plt.show()"""

#3.实现K-means算法

def findClosestCentroids(x,center):
    samples = x.shape[0] #样本数量:300
    centroids = np.zeros(samples)
    for i in range(samples):
        temp = x[i,:] - center
        c_temp = np.linalg.norm(x=temp, ord=2,axis=1,keepdims=False)
        #axis=1表示对矩阵的每一行求范数
        #keepdims=False表示结果不保留二维特性
        centroids[i] = np.argmin(c_temp)
        #找出最大值的索引
    return centroids


def computeCentroids(x,classes):
    result = np.zeros((3,2))
    for i in range(3):
        index = np.where(classes == i)
        temp_x = x[index]
        result[i,:] = np.sum(temp_x,axis=0)/temp_x.shape[0]
    return result




def K_means(x,centroids):
    #不放回地取样k个样本作为簇中心
    for i in range(10):
        classes = findClosestCentroids(x,centroids)
        centroids = computeCentroids(x, classes)
    return centroids
    # [[2.42830111 3.15792418]
    #  [5.81350331 2.63365645]
    # [7.11938687
    # 3.6166844]]

k=3
centroids = np.array([[3, 3], [6, 2], [8, 5]])
centroids = K_means(x,centroids)
classes = findClosestCentroids(x,centroids)

colors = ['r', 'g', 'b']
markers = ['o','+','x']
for i in range(3):
    index_x = np.where(classes==i)
    xi = x[index_x]
    plt.scatter(xi[:,0],xi[:,1],c=colors[i],marker=markers[i])
plt.show()

任务二：使用K-means算法进行图像压缩

方法一：使用自己定义的函数

10次随机初始化，10次迭代

100次随机初始化，10次迭代

10次随机初始化，100次迭代

import numpy as np
from matplotlib import pyplot as plt
import matplotlib.image as mpimg

#1.读取数据
dt = mpimg.imread(r"E:\机器学习\吴恩达\data_sets\bird_small.png") #(128, 128, 3)

#2.可视化
"""plt.imshow(dt) # 显示图片
plt.axis('off') # 不显示坐标轴
plt.show()"""


#3.进行K-means
#随机初始化簇中心
def kMeansInitCentroids(x, k):
    randidx = np.random.permutation(x) #随机排列
    centroids = randidx[:k, :] #选前K个
    return centroids



#进行簇分配cluster assignment
def findClosestCentroids(x,center):
    samples = x.shape[0] #样本数量:300
    centroids = np.zeros(samples)
    sum = 0
    for i in range(samples):
        temp = x[i,:] - center
        c_temp = np.linalg.norm(x=temp, ord=2,axis=1,keepdims=False)
        #axis=1表示对矩阵的每一行求范数
        #keepdims=False表示结果不保留二维特性
        centroids[i] = np.argmin(c_temp)
        #找出最大值的索引
        sum += np.sum(x[i,:]-center[int(centroids[i])])
    return centroids,sum


#计算簇中心
def computeCentroids(x,classes,k):
    result = np.zeros((k,x.shape[1]))
    for i in range(k):
        index = np.where(classes == i)
        temp_x = x[index]
        result[i,:] = np.sum(temp_x,axis=0)/temp_x.shape[0]
    return result

def k_means(x,k,max_init,max_iters):
    cost = INF
    for j in range(max_init):
        center = kMeansInitCentroids(x, k)
        for i in range(max_iters):
            classes,temp_sum = findClosestCentroids(x, center)
            center = computeCentroids(x,classes,k)
            if temp_sum < cost:
                cost = temp_sum
                result = center
    return result

INF = 1e10
x = dt.reshape(-1,3)
K = 16
MAXIMUM_iterations = 10
MAXIMUM_init = 100
centroids = k_means(x,K,MAXIMUM_init,MAXIMUM_iterations)
classes,p = findClosestCentroids(x, centroids)

presssed = np.zeros(x.shape)
for i in range(len(centroids)):
    presssed[classes==i] = centroids[i]

presssed = presssed.reshape(dt.shape)
plt.imshow(presssed)
plt.show()

方法二：调包：使用MiniBatchKMeans

import numpy as np
from matplotlib import pyplot as plt
import matplotlib.image as mpimg
from sklearn.cluster import MiniBatchKMeans


#1.读取数据
dt = mpimg.imread(r"E:\机器学习\吴恩达\data_sets\bird_small.png") #(128, 128, 3)

#2.可视化
"""plt.imshow(dt) # 显示图片
plt.axis('off') # 不显示坐标轴
plt.show()"""

#3.使用MiniBatchKMeans聚类
K = 16
temp = dt.reshape(-1,3)
kmeans = MiniBatchKMeans(K)
kmeans.fit(temp) #先fit
pressed = kmeans.cluster_centers_[kmeans.predict(temp)] #再predict，然后再分配标记的聚类中心的RGB编码
pressed = pressed.reshape(dt.shape)
plt.imshow(pressed)
plt.show()