Python实现非正太分布的异常值检测方式

# -*- coding: utf-8 -*-
"""
Created on Tue Apr 30 10:52:37 2019
@author: chen_lib
"""
 
import pandas as pd
catering_sale = 'D:/Users/chen_lib/Desktop/ceshi.csv' #读取历史数据
datax = pd.read_csv(catering_sale) #读取数据
#取出不是昨天的数据
data = datax.loc[datax['orderdate'] != datetime][:]
'''
import time
## yyyy-mm-dd格式
print (time.strftime("%Y-%m-%d"))
'''
#时间减一天
import datetime
datetime = (datetime.datetime.now()+datetime.timedelta(days=-1)).strftime("%Y-%m-%d")
 
 
#保存基本统计量，将常见的统计信息保存为数据框
statistics = data.describe() 
#添加行标签 计算出每个指标的上线下线和四分位间距
statistics.loc['IQR'] = statistics.loc['75%']-statistics.loc['25%'] #四分位数间距
statistics.loc['UP'] = statistics.loc['75%'] + 1.5*statistics.loc['IQR'] #上限
statistics.loc['DAWN'] = statistics.loc['25%'] - 1.5*statistics.loc['IQR']#下限
#取出data的列名
columns = data.columns.values.tolist()
 
 
'''取出要比较的数值,放在统计信息表'''
a = data.loc[data['orderdate'] == datetime][columns[1]]#取出第一列
b = data.loc[data['orderdate'] == datetime][columns[2]]#取出第二列
statistics.loc['res'] = [a[1],b[1]]#取出需要比较的当天的数据 放入统计信息中
  
 
'''循环取出结果是否满足要求''' 
ret = [] 
for i in range(2):
  res = statistics.loc['res'][i]
  max = statistics.loc['UP'][columns[i+1]]#最大值
  min = statistics.loc['DAWN'][columns[i+1]]#最小值
  '''
  #重建三个值的索引，以便比较大小
  res.index = ['ordernum']
  max.index = max['ordernum']
  min.index = min['ordernum']
  #判断异常值，若大于最大值或者小于最小值则抛出结果为1
  '''
  result1 = res>max
  result2 = res<min
  if result1 =='False' or result2 == 'False':
    ret.append([columns[i+1],1])
  else: 
    ret.append([columns[i+1],0])
  df = pd.DataFrame(ret)
  
	#将文件写入excel表中
df.to_excel("d:/Users/chen_lib/Desktop/ceshi.xlsx",sheet_name="total",index=False,header=False)