代码笔记，仅供参考

---

回厂日期预测

前言，对不同客户的下一次返厂时间进行预测，大多数客户的返厂次数不足10次，仅有少量客户返厂次数大于30次。

平均值法预测(python)

# -*- coding: utf-8 -*-import pymysql
import time
import numpy as np
import pandas as pdclass CarTest:def __init__(self):self.db = pymysql.connect(host = '127.0.0.1',port = 3306,user = 'root',password = '19970928',database = 'datacup',charset = 'utf8')self.cur = self.db.cursor()self.avertime = Noneself.datetframe = Noneself.preTime = {}self.Accuracy = 0def getDate(self):with open("./data/car_srv_train.csv", 'r') as f:data = f.readline()data = f.readline()print(type(data))while (data):
#                data = self.timeDeal(data)self.mysqlInteractive(data)data = f.readline()def timePre(self):dictTest = {}for key, value in self.dataframe:if key not in dictTest:dictTest[key] = [value]else:dictTest[key].append(value)
#            print(dictTest[key])
#        print(dictTest)for key in dictTest:if len(dictTest[key]) > 5:
#                print(dictTest[key])sortTemp = dictTest[key]sortTemp.sort(reverse = True)
#                print(sortTemp)
#                print(type(sortTemp))averTimeInterval = (sortTemp[1] - sortTemp[-1])/(len(dictTest[key])-1)divTime = sortTemp[1] - sortTemp[2]preTime = (averTimeInterval + divTime)/2self.preTime[key] = [preTime]self.preTime[key].append(sortTemp[0] - sortTemp[1])self.preTime[key].append(sortTemp[1])self.preTime[key].append(sortTemp[0])#1:预测间隔时间，2：真实间隔时间，3：倒数第二天， 4：倒数第一天def errorTest(self):#设置预测时间在3个工作日内为预测正确trueNum = 0countNum = 0preDictTemp = self.preTimefor key in preDictTemp: #value为引用数据类型，可能不用key,value接收循环countNum +=1#            print(preDictTemp[key][1]-preDictTemp[key][0]) 
#            print(type(preDictTemp[key][1]-preDictTemp[key][0])) if np.abs((preDictTemp[key][1]-preDictTemp[key][0]).days) < 3:trueNum += 1self.Accuracy = trueNum/countNumdef mysqlInteractive(self, data):dataList = data.split(",")[0:3]
#        print(dataList)sqlLine = "insert into CarTest(ordername, username, ordertime) \values(%s, %s, %s);"try:self.cur.execute(sqlLine, dataList)self.db.commit()print("提交成功...")    except Exception as e:self.db.rollback()print('错误信息:', e) def getMysqData(self):sqlLine = "select username, ordertime from CarTest \order by username,ordertime;"      
#        sqlLine2 = """select username, DATEDIFF(max(ordertime), min(ordertime))/count(*) as avertime, count(*) as countnum from 
#                    cartest group by username HAVING COUNT(*)>=3;"""self.cur.execute(sqlLine)self.dataframe = self.cur.fetchall()
#        self.db.commit()
#        self.cur.execute(sqlLine2)
#        self.avertime = self.cur.fetchall()
#        self.db.commit()def main(self):
#        self.getDate()self.getMysqData()self.timePre()self.errorTest()self.cur.close()self.db.close()if __name__ == '__main__':#获取开始的时间戳start = time.time()dataOutLine = CarTest()dataOutLine.main()print("准确率为:", dataOutLine.Accuracy)#获取结束的时间戳end = time.time()print('执行时间:%.2f' % (end-start))

输出：

准确率为: 0.016651248843663275
执行时间:23.67

好吧，准确率低的可以。

时间序列预测(R语言)

library(forecast)#读取数据getwd()
setwd("C:/Users/goatbishop/Desktop/data")
car_srv_train <- read.csv("car_srv_train.csv", header = T, stringsAsFactors = F)#简单查看数据
head(car_srv_train)
dim(car_srv_train)str(car_srv_train)
car_srv_train$ORDERDATE <- as.Date(car_srv_train$ORDERDATE)test0624car <- car_srv_train
head(diff(test0624car$ORDERDATE))test0624car <- test0624car[order(test0624car$CUST_ID, test0624car$ORDERDATE, decreasing = F), ]
#head(test0624car[,c(1:4)])table(test0624car$CUST_ID)
innames <- names(which(table(test0624car$CUST_ID) >= 3))test0624car3 <- test0624car[which(test0624car$CUST_ID %in% innames), ]
tablenum <- table(test0624car3$CUST_ID)
length(tablenum)
newtest0624car <- test0624car3[-cumsum(tablenum), ]
dim(test0624car3)
dim(newtest0624car)test0624diff <- tapply(test0624car3$ORDERDATE, test0624car3$CUST_ID, diff)
dfg <- unlist(test0624diff)
dftest <- cbind(newtest0624car, dfg)head(dftest[, c(1:3, 13)], 10)
write.csv(dftest, "dftest.csv")#####时间序列#####innames2 <- names(which(table(test0624car$CUST_ID) >= 30))Pre <- c()
real <- c()
for (item in c(1:length(innames2))) {tempdf <- dftest$dfg[which(dftest$CUST_ID == innames2[item])]temparima <- auto.arima(tempdf[-length(tempdf)])preout <-forecast(temparima, 1)real <- c(real, tempdf[length(tempdf)])Pre <- c(Pre, preout$mean)}errorPre <- real - Pre
write.csv(Pre, "Pre.csv")###精确度###a1 = 0
b1 = 0
for (item in c(1:length(errorPre))) {b1 = b1 + 1if (abs(errorPre[item]) < 10) {a1 = a1 + 1}
}(accer <- a1/b1)write.csv(errorPre, "errorPre.csv")
write.csv(accer, "accer.csv")which(abs(errorPre) < 2)tempdf <- dftest$dfg[which(dftest$CUST_ID == innames2[4])]
plot(tempdf, type = 'o', main = "时序图")temparima <- auto.arima(tempdf[-length(tempdf)])
pretest <- forecast(temparima)acf(pretest$residuals)
write.csv(pretest$residuals, "pretest_residuals.csv")