半监督kmeans聚类

根据西瓜书的算法实现半监督kmeans聚类

import numpy as np
import pandas as pd


class kmean_half:
    def __init__(self, data, label):
        self.loaddata = data
        self.label = label

    def data_load(self):
        # 数据划分为有标签数据和无标签数据
        label_data = self.loaddata[self.loaddata[self.label].isnull() == False]
        unlabel_data = self.loaddata[self.loaddata[self.label].isnull() == True]
        return label_data, unlabel_data

    # 初始化聚类中心
    def clu_center_ini(self, label_data):
        """
        :param label_data: 有标签数据
        :return:
        """
        # 获取所有类别
        labels = label_data[self.label].unique()
        # 计算各聚类中心
        clu_center = []
        for lab in labels:
            df = label_data[label_data[self.label] == lab]
            df = df.drop(self.label, axis=1)
            df = np.array(df)
            clu_center.append(np.mean(df, axis=0))
        return clu_center, labels

    # 有标记样本划分到各簇中
    def label_sample_clu(self, data, labels: list):
        """

        :param data: 有标记样本数据
        :param labels: 类别列表（去重）
        :return:
        """

        label_id = data[self.label].values
        data = data.drop(self.label, axis=1)
        data = np.array(data)
        sample_list = []
        for i in range(len(labels)):
            sample_list.append([])
        for i in range(len(data)):
            id = np.where(labels == label_id[i])
            sample_list[id[0][0]].append(data[i])
        return sample_list

    # 计算新的聚类中心
    def new_center(self, label_list):
        """
        :param label_list: 样本簇列表
        :return:
        """
        label_list = np.array(label_list)
        for i in range(len(label_list)):
            label_list[i] = np.array(label_list[i])
        center = []
        for i in range(len(label_list)):
            a = np.mean(label_list[i], axis=0)
            center.append(a)
        return center

    # 计算两个向量之间的欧氏距离
    def sample_center_distance(self, x, y):
        return np.linalg.norm(x - y)

    # 样本簇更新
    def new_clu(self, new_center, data):
        sample_list = []
        for i in range(len(new_center)):
            sample_list.append([])
        df = data.drop(self.label, axis=1)
        df = np.array(df)
        for i in range(len(df)):
            tmp = []
            for j in range(len(new_center)):
                tmp.append(self.sample_center_distance(df[i], np.array(new_center[j])))
            id = np.where(tmp == min(tmp))
            sample_list[id[0][0]].append(df[i])
        return sample_list

    def run(self, errors):
        # 加载数据
        label_data, unlabel_data = self.data_load()
        # 初始化聚类中心
        clu_center, labels = self.clu_center_ini(label_data)
        # 有标记样本划分到簇中
        label_sample_list = self.label_sample_clu(label_data, labels)
        # 无标记样本划分到簇中
        sample_list = self.new_clu(clu_center, unlabel_data)
        # 两种样本合并
        samplelist = []
        for i in range(len(labels)):
            samplelist.append(label_sample_list[i] + sample_list[i])
        # 计算新的聚类中心
        center = self.new_center(samplelist)
        # 计算新的聚类中心与旧聚类中心的距离
        e = self.sample_center_distance(np.array(center), np.array(clu_center))
        # 以下是常规kmeans聚类
        while e > errors:
            # 根据新的聚类中心重新划分样本
            sample_list = self.new_clu(center, unlabel_data)
            # 计算新的聚类中心
            new_center = self.new_center(sample_list)
            # 计算新的聚类中心与旧聚类中心的距离
            e = self.sample_center_distance(np.array(center), np.array(new_center))
            center = new_center
        return sample_list, labels