欢迎您访问程序员文章站本站旨在为大家提供分享程序员计算机编程知识!
您现在的位置是: 首页  >  IT编程

如何使用python 和excel 对100M左右的数据进行保存和简单分析

程序员文章站 2022-04-14 09:30:51
老师给了一个100m左右的数据(.csv 格式),然后让我们用python 和excel 进行分析。 具体要求和代码: t2.py import sqlite3 import pandas a...

老师给了一个100m左右的数据(.csv 格式),然后让我们用python 和excel 进行分析。

具体要求和代码:

t2.py

import sqlite3
import pandas as pd

####   main       ##################################################
##分为两个步骤,先是对数据读取后进行处理
##然后将数据放入数据库
if __name__ == '__main__':
    ##三个文件路径
    ###########  对数据读取后进行处理   ############################################################
    filename = ['./data/opendata_order201608_afterclear.csv', './data/opendata_order201609_afterclear.csv',
                './data/opendata_order201610_afterclear.csv']
    ##对每个文件进行处理
    for index in range(3):
        ##一行行处理数据,将所需要增加的都放在后面,不影响结果,但是操作更加方便
        rfile = open(filename[index], "r")
        # 用来临时保存数据的文件,opendata_order201608.csv
        targetfile = filename[index][7:27] + '.csv'
        wfile = open(targetfile, "w")
        # print(targetfile)
        linecount = 0
        dataline = rfile.readline()  # 读取一行数据
        while dataline:
            linecount += 1
            # 分割数据,用来接下来的处理
            datalist = dataline.split(",")
            # print(datalist)
            ##对第一行,即header进行处理,增加所要求的内容
            if (linecount == 1):
                datalist.append('month')
                datalist.append('year')
                datalist.append('area_name')
                datalist.append('brand_code')
                # print(datalist)
            else:
                # 对具体内容进行处理,增加所要求的内容
                month = (datalist[0][4:6])
                #
                year = (datalist[0][0:4])
                # print(year)
                area_name = 'qx' + datalist[1][-3:]
                # print(area_name)
                brand_code = datalist[3][0:6]
                # print(brand_code)
                datalist.append(month)
                datalist.append(year)
                datalist.append(area_name)
                datalist.append(brand_code)

                # print(datalist)
            ##将内容整理好,放入临时文件
            linecontent = ""
            rolcount = 0
            for col in datalist:
                rolcount += 1
                if (rolcount == 1):
                    linecontent += col.strip()
                else:
                    linecontent += "," + col.strip()
            # 一行行写入
            wfile.write(linecontent + "\n")
            # print(linecontent)
            dataline = rfile.readline()
        print("数据行数=", linecount)
        rfile.close()
        wfile.close()

        ##########       将数据放入数据库             ###################################

        # 连接数据库
        conn = sqlite3.connect("./db/saledata.db")
        ##采用pandas 的一个函数,可以很方便的直接将数据从文件中写入到数据库
        df = pd.read_csv(targetfile)
        df.to_sql(filename[index][7:27], conn, if_exists='append', index=false)

t3.py

import sqlite3

# 连接数据库
conn = sqlite3.connect('./db/saledata.db')
# 查询语句
query_sql = '''
select
opendata_order201608.kind,
opendata_order201608.area_name,
sum(opendata_order201608.purch_tax_amt)
from
opendata_order201608
group by
opendata_order201608.kind,
opendata_order201608.area_name
'''
##进行查询
query = conn.execute(query_sql)
##将查询结果写入
filename = './txt/areakind.txt'
fd = open(filename, 'w', encoding='utf-8')
print('烟的种类', '县区', '销售利润', file=fd, sep=',')
for kind in query:
    print(kind[0], kind[1], kind[2], file=fd, sep=',')
fd.close()
# print(query[1][1])

t4.py

import sqlite3

# 连接数据库
conn = sqlite3.connect('./db/saledata.db')
###############################################
##对每个月的查询分开处理,先是8月
# 查询语句
query_sql = '''
select
opendata_order201608.cust_code,avg(purch_tax_amt)
from
opendata_order201608
group by
opendata_order201608.cust_code
'''
##查询
query = conn.execute(query_sql)
##将查询结果进行写入
filename = './txt/t4.txt'
num = 0
fd = open(filename, 'w', encoding='utf-8')
print('月份', '商家', '销售利润', file=fd, sep=',')
for kind in query:
    num += 1
    print('8月', kind[0], kind[1], file=fd, sep=',')
print('8月商户数量:', num)
###################     9月     ######################################
query_sql = '''
select
opendata_order201609.cust_code,avg(purch_tax_amt)
from
opendata_order201609
group by
opendata_order201609.cust_code
'''
query = conn.execute(query_sql)

filename = './txt/t4.txt'
num = 0
fd = open(filename, 'a', encoding='utf-8')
# print('月份','商家','销售利润',file=fd,sep=',')
for kind in query:
    num += 1
    print('9月', kind[0], kind[1], file=fd, sep=',')
print('9月商户数量:', num)
###################       10月    ####################################
query_sql = '''
select
opendata_order201610.cust_code,avg(purch_tax_amt)
from
opendata_order201610
group by
opendata_order201610.cust_code
'''
query = conn.execute(query_sql)

filename = './txt/t4.txt'
num = 0
fd = open(filename, 'a', encoding='utf-8')
# print('月份','商家','销售利润',file=fd,sep=',')
for kind in query:
    num += 1
    print('10月', kind[0], kind[1], file=fd, sep=',')
print('10月商户数量:', num)

fd.close()
t5.py
import sqlite3

##连接数据库
conn = sqlite3.connect('./db/saledata.db')

##对每个月的查询分开处理,先是8月##############################################
query_sql = '''
select
opendata_order201608.item_code,
sum(opendata_order201608.purch_qty),
sum(opendata_order201608.purch_tax_amt)/sum(opendata_order201608.purch_qty)
from
opendata_order201608
group by
opendata_order201608.item_code
'''
query = conn.execute(query_sql)

filename = './txt/t5.txt'
num = 0
fd = open(filename, 'w', encoding='utf-8')
print('月', '规格', '数量', '单价', file=fd, sep=',')
for kind in query:

    if (kind[1] == 0):
        continue
    num += 1
    print('8月', kind[0], kind[1], kind[2], file=fd, sep=',')
print(num)
####################   9月    ############################################
query_sql = '''
select
opendata_order201609.item_code,
opendata_order201609.purch_qty,
opendata_order201609.purch_tax_amt/opendata_order201609.purch_qty
from
opendata_order201609
group by
opendata_order201609.item_code
'''
query = conn.execute(query_sql)

filename = './txt/t5.txt'
num = 0
fd = open(filename, 'a', encoding='utf-8')
# print('月','规格','数量','单价',file=fd,sep=',')
for kind in query:

    if (kind[1] == 0):
        continue
    num += 1
    print('9月', kind[0], kind[1], kind[2], file=fd, sep=',')
print(num)
#######################    10月  ##################################################
query_sql = '''
select
opendata_order201610.item_code,
opendata_order201610.purch_qty,
opendata_order201610.purch_tax_amt/opendata_order201610.purch_qty
from
opendata_order201610
group by
opendata_order201610.item_code
'''
query = conn.execute(query_sql)

filename = './txt/t5.txt'
num = 0
fd = open(filename, 'a', encoding='utf-8')
# print('月','规格','数量','单价',file=fd,sep=',')
for kind in query:

    if (kind[1] == 0):
        continue
    num += 1
    print('10月', kind[0], kind[1], kind[2], file=fd, sep=',')
print(num)
fd.close()

t6.py

import numpy
import sqlite3

# 连接数据库
conn = sqlite3.connect('./db/saledata.db')
##由于需要知道每一个商家的销售额占总销售额的比例,所以需要查询两次
# 第一次   每一个商家的销售额
query_sql = '''
select
opendata_order201608.cust_code,
sum(opendata_order201608.purch_tax_amt)
from
opendata_order201608
group by
opendata_order201608.cust_code
order by
sum(opendata_order201608.purch_tax_amt) desc


'''
query = conn.execute(query_sql)
# 第二次   总的销售额
query_sql_total = '''
select
sum(opendata_order201608.purch_tax_amt)
from
opendata_order201608
'''
query_total = conn.execute(query_sql_total)
total_sum = query_total.fetchone()[0]
##      写入     ####################################################
filename = './txt/t6.txt'
num_total = 0  # 总的商家数
fd = open(filename, 'w', encoding='utf-8')
print('商户', '销售金额', '占比', file=fd, sep=',')
zhanbi_rate = 0  # 销售收入占比率
flag_num = 0  # 当超过80时,商家的个数
flag = 0  # 用来标记第一次超过80%
for kind in query:
    num_total += 1
    zhanbi = round((kind[1]) / (total_sum) * 100, 5)
    # 百分比    保留五位小数
    zhanbi_rate += zhanbi
    if zhanbi_rate > 80 and flag == 0:
        flag_num = num_total
        print('当商户的销售占比率超过80%,商户的数量:    ', flag_num)
        flag = 1
    print(kind[0], kind[1], zhanbi, file=fd, sep=',')
print('总的商户数量   :', num_total)
print('商户占比率   ', round(flag_num / num_total, 5))

t7.py

import sqlite3
import numpy as np

# 连接数据库
conn = sqlite3.connect('./db/saledata.db')
#####   平均值      ##############################
query_sql = '''
select
opendata_order201608.cust_code,
avg(opendata_order201608.purch_tax_amt)

from
opendata_order201608
group by
opendata_order201608.cust_code
order by
opendata_order201608.cust_code asc
'''
query = conn.execute(query_sql)
######      方差              ############################
query_sql_per = '''
select
opendata_order201608.cust_code,
opendata_order201608.purch_tax_amt
from
opendata_order201608
order by
opendata_order201608.cust_code asc
'''
query_per = conn.execute(query_sql_per)

purch_per = []  # 用来保存每一个商家的销售额
fangcha = {}  # 用来保存每一个商家的销售额的方差
cust_code = 1  # 用来保存每一个商家的cust_code
last_custcode = 0  # 用来保存最后一个商家的cust_code
for i in query_per.fetchall():
    # 如果当前cust_code不等于之前的cust_code,则说明上一个商家的记录结束,进行处理
    if i[0] != cust_code:
        # print(purch_per)
        # 计算方差
        cust_fangcha = round(np.std(purch_per, ddof=1), 5)
        # print(cust_fangcha)
        # 用字典保存方差
        fangcha[cust_code] = cust_fangcha
        # 更新 cust_code
        cust_code = i[0]
        ##将这个[]清空
        purch_per = []
    # 保存销售额
    purch_per.append(i[1])
    last_custcode = i[0]
###在之前的操作中,没有计算最后一个商家的方差,这里加上
last_fangcha = round(np.std(purch_per, ddof=1), 5)
fangcha[last_custcode] = last_fangcha
############    写入      ##############################
filename = './txt/t7.txt'
fd = open(filename, 'w', encoding='utf-8')
print('商家', '平均', '方差', file=fd, sep=',')
# fnagchanum=0
for kind in query:
    print(kind[0], round(kind[1], 5), fangcha[kind[0]], file=fd, sep=',')
    # fnagchanum += 1
fd.close()

t8.py

import sqlite3

# 连接数据库
conn = sqlite3.connect('./db/saledata.db')
# 查询语句
query_sql = '''
select
opendata_order201608.cust_code,
sum(opendata_order201608.purch_tax_amt)
from
opendata_order201608
group by
opendata_order201608.cust_code
order by
sum(purch_tax_amt) desc
'''
# 查询
query = conn.execute(query_sql)

# 遍历一遍,计算商家数量
num_cust = 0
for kind in query:
    num_cust += 1
# 一个很有趣的现象,如果查询结果被遍历一边,然后query就为[],所以需要再遍历一遍
query_again = conn.execute(query_sql)
numagain = 0
# 每个等分中 商家的销售额
sumper = [0 for i in range(5)]
# print(query.fetchall())
for per in query_again:
    index = int(numagain / (num_cust / 5))
    # print(index)
    sumper[index] += per[1]
    # print('-'*16)
    numagain += 1
# 保存所要求的倍数
rate = []
for index in range(5):
    # 计算倍数,写入list
    rate.append(round(sumper[0] / sumper[index], 3))
    print(rate[index])
# print(sumper)

##写入
filename = './txt/t8.txt'
fd = open(filename, 'w', encoding='utf-8')
print('等分', '比值', file=fd, sep=',')
for index in range(5):
    print(index, rate[index], file=fd, sep=',')
fd.close()

t9.py

import sqlite3
#在这一道题中,我想要看一下每一类烟的平均单价与销售数量,看看有什么联系
##连接数据库
conn = sqlite3.connect('./db/saledata.db')

##对每个月的查询分开处理,先是8月##############################################
query_sql = '''
select
opendata_order201608.brand_code,
sum(opendata_order201608.purch_tax_amt)/sum(opendata_order201608.purch_qty),
sum(opendata_order201608.purch_qty)
from
opendata_order201608
group by
opendata_order201608.brand_code
order by
sum(opendata_order201608.purch_tax_amt)/sum(opendata_order201608.purch_qty) desc
'''
query = conn.execute(query_sql)

filename = './txt/t9.txt'
num = 0
fd = open(filename, 'w', encoding='utf-8')
print( '品牌', '单价', '数量', file=fd, sep=',')
for kind in query:
    if (kind[1] == 0):
        continue
    num += 1
    print( kind[0], kind[1], kind[2], file=fd, sep=',')
print(num)