#WSS未知異常預測第一種演算法實現——kmeans改進版聚類演算法
import numpy as np
from scipy.spatial.distance import cdist
from sklearn.cluster import KMeans
import pandas as pd

class kmeans_optimization:

    def __init__(self, data):  # 傳入一個二維的陣列numpy陣列為所需要檢測的資料data
        self.data =np.array(data)

    #結合實際資料實現的改進版本聚類方法
 
    def run(self):
        # 首先進行k_means聚類表示
        self.kmeans = KMeans(n_clusters=2, n_init=100)
        self.kmeans.fit(self.data)
        kmeans_pre = self.kmeans.labels_
        self.r1 = self.data[kmeans_pre == 0]
        self.r2 = self.data[kmeans_pre == 1]
        self.kmeans_center = self.kmeans.cluster_centers_  # 獲取聚類中心
 

        # 計算聚類的結果到各自聚類中心的距離
        self.distance1 = cdist(self.r1, [self.kmeans_center[0]])
        self.distance2 = cdist(self.r2, [self.kmeans_center[1]])
        # 進行k_means聚類合理性判斷
        if len(self.r1) < 0.1 * len(self.data) or len(self.r2) < 0.1 * len(self.data):
            if len(self.r1) < len(self.r2):
 
                self.errordata = self.r1
                self.normaldata = self.r2
            else:
                self.errordata = self.r2
                self.normaldata = self.r1

        else:
            if self.distance1.max() < self.distance2.max():
                self.distance = self.distance2
                self.distance_1 = self.distance1
                self.errordata1 = self.r2
                self.normaldata1 = self.r1
            else:
                self.distance = self.distance1
                self.distance_1 = self.distance2
                self.errordata1 = self.r1
                self.normaldata1 = self.r2

            #定義輸出異常距離輸出的閾值大小
            self.threshold = self.distance.mean() + 3 * self.distance1.std()

            self.error_ind1 = []
            self.normal_ind1 = []
            for (i, v) in enumerate(self.distance):
                if v > self.threshold:
                    self.error_ind1.append(i)
                else:
                    self.normal_ind1.append(i)
            self.errordata = self.errordata1[self.error_ind1]                     #得到異常的資料集合
            self.normaldata2 = self.errordata1[self.normal_ind1]
            self.normaldata = np.vstack((self.normaldata1, self.normaldata2))     #得到正常的資料集合

        #索引查詢和返回
        self.errorindex = []
        for (i, v) in enumerate(self.errordata):
            for (j, v1) in enumerate(self.data):
                if (v1 == v).all():
                    self.errorindex.append(j)
        self.index = [i for i in range(len(self.data))]
        self.normaldataindex = [i for i in self.index if i not in self.errorindex]

        return self.errordata,self.normaldata,self.errorindex,self.normaldataindex

if __name__ == '__main__':
    x = np.random.normal(1, 0.5, size=(20000, 69))
    print(x)
    y = np.random.normal(5, 0.5, size=(10, 69))
    print(y)
    y1 = np.random.normal(7, 1, (15, 69))
    y2 = np.random.normal(10, 1, (10, 69))
    z = np.vstack((x, y, y1, y2))
    z = np.array(z)
    print(z)
    z = pd.read_excel("finaldata.xlsx")
    z = z.iloc[:, :69]
    z = np.array(z)
    k = kmeans_optimization(z)
    print(k.run())