基本原理

代码实现

先给出完整代码，再分别说明

#-*- coding:utf-8

import traceback
from sklearn.cluster import KMeans
import numpy as np
import matplotlib.pyplot as plt

'''
函数名:draw_original
功能:根据样本和真相画图
@X:样本
@Y:真相
'''
def draw_original(X, Y):
    try:

        plt.subplot(1,1,1)
        plt.scatter(X[:, 0], X[:, 1], c=Y)

        plt.title("original clusters")
        plt.xlabel("Feature1")
        plt.ylabel("Feature2")
        plt.show()

    except Exception,e:
        print traceback.print_exc()

'''
函数名: drawing_n_clusters
功能: 根据聚类列表,聚类中心画图
@cluster_list: 聚类列表，每个元素是一个聚类。这个聚类样本的列表，注意是列表
@centroids: 聚类中心，矩阵形式存储
'''
def drawing_n_clusters(cluster_list, centroids):
    try:
        n_clusters = len(cluster_list)
        k_clusters = [np.array(cluster) for cluster in cluster_list]  # 每一个类别以矩阵形式存储

        plt.subplot(1, 1, 1)
        color_list = ['b', 'g', 'r', 'c', 'm', 'y', 'k', 'w']
        for idx, cluster in enumerate(k_clusters):
            label_str = "cluster" + str(idx)
            plt.scatter(cluster[:, 0], cluster[:, 1], c=color_list[idx], label=label_str)

        plt.scatter(centroids[:, 0], centroids[:, 1], marker='+', color='y', s=200)

        title = str(n_clusters) + " clusters"
        plt.title(title)
        plt.xlabel("Feature1")
        plt.ylabel("Feature2")
        plt.legend(loc=2)

        plt.show()

    except Exception, e:
        print traceback.print_exc()

'''
函数名: draw_error
功能: 画出不同K时的损失和
@k_list: K列表
@error_list: 每个K下的损失和
'''
def draw_error( k_list, error_list ):
    try:
        # 画框设置
        plt.figure(figsize=(8, 5), dpi=80) # 创建图像
        plt.subplot(1,1,1)

        # 画点
        plt.plot(k_list, error_list, marker='o', c = 'blue')

        # 加标题
        plt.title("Total Error vs. # of Clusters")
        # 加坐标轴
        plt.xticks(k_list)
        plt.xlabel("k")
        plt.ylabel("total squared error")

        # 显示
        plt.show()

    except Exception,e:
        print traceback.print_exc()

'''
-------------------------------------------------------------------------------
'''

'''
函数名：load_dataset
功能：加载数据集：
@input_path:输入路径
@X: 样本列表，每个样本也以列表形式存储
@Y: 原始标签列表，以列表形式存储
'''
def load_dataset( input_path ):
    try:

        X = []
        Y = []

        infile = open(input_path, "r")

        for line in infile:
            data = line.rstrip('\r\n').split('\t')
            x = []
            y = []
            y.append(int(data[0]))
            x.append(float(data[1]))
            x.append(float(data[2]))

            X.append(x)
            Y.append(y)

        infile.close()
        return X, Y

        print "[INFO]: load_dataset is finished!"
    except Exception,e:
        print traceback.print_exc()



'''
函数名：training
功能：训练kmeans聚类器，初始点的选择采用kmeans++，对于K的训练迭代多次，返回最优值的聚类结果
@X: 样本-矩阵形式，均以向量的形式保存
@K: 聚类数量
@label: 返回每个样本的训练标签
@loss: means square均方误差
@centroids: 聚类中心
'''
def training( X, K ):
    try:

        kmeans = KMeans(n_clusters=K).fit(X)

        label = kmeans.labels_
        loss = kmeans.inertia_
        centroids = kmeans.cluster_centers_

        return label, loss, centroids

    except Exception,e:
        print traceback.print_exc()

'''
函数名:get_clusters
功能: 根据样本和聚类结果，获得每个聚类
@X: 样本-矩阵形式
@label: 样本标签-矩阵形式
@K: 聚类数量
@cluster_list: 类别
'''
def get_clusters(X, label, K):
    try:

        cluster_list = [ [] for x in range(K) ] # 每一个聚类用一个列表存。每个列表存这个聚类的样本向量
        idx = 0
        len_label = len(label)
        while idx < len_label:
            cluster_list[label[idx]].append(X[idx])
            idx += 1

        return cluster_list

    except Exception,e:
        print traceback.print_exc()

'''
-------------------------------------------------------------------------------
'''
def find_K():
    try:

        INPUT_PATH = "../data/4k2_far.txt"
        OUTPUT_PATH_K = "../output/test_for_4k2/inertia.txt"

        _X, _Y = load_dataset(INPUT_PATH)

        X = np.array(_X)
        #Y = np.array(_Y)

        error_list = []
        outfile = open(OUTPUT_PATH_K, "w")
        for K in range(1,21):
            _, loss, _ = training(X, K)
            error_list.append(loss)
            line = "K=" + str(K) + "," + str(loss)
            outfile.write(line + '\n')
        outfile.close()

        k_list = [ k for k in range(1,21) ]
        draw_error(k_list, error_list)

        print "[INFO]: find_K is finished!"
    except Exception,e:
        print traceback.print_exc()

def main():
    try:
        INPUT_PATH = "../data/4k2_far.txt"

        _X, _Y = load_dataset(INPUT_PATH)

        X = np.array(_X)
        Y = np.array(_Y)

        K = 2
        label, loss, centroids = training(X, K)

        cluster_list = get_clusters(X, label, K)
        drawing_n_clusters(cluster_list, centroids)

    except Exception,e:
        print traceback.print_exc()

if __name__ == '__main__':
    #main()
    find_K()