当前位置:主页 > 软件编程 > Python代码 >

基于Python实现对比Exce的工具

时间:2022-07-10 09:40:05 | 栏目:Python代码 | 点击:

目的:设计一个应用GUI用于对比两个Excel文件

思路

1.参数

(默认新旧文件列名一致)

2.效果

3.实现

配置

import pandas as pd
from openpyxl import load_workbook
#选择文件路径
path=r"C:\Users\小管同学\Desktop\Migration_Data_Compari\对比文件.xls" #input("选择文件路径:")
TargetPath=r"C:\Users\小管同学\Desktop\Migration_Data_Comparison_Tool\目标文件\对比结果.xlsx"
DATA_ODS=pd.read_excel(r"C:\Users\小管同学\Desktop\Migration_Data_Comparison_Tool\对比文件.xls",sheet_name="ODS")
DATA_DWH=pd.read_excel(r"C:\Users\小管同学\Desktop\Migration_Data_Comparison_Tool\对比文件.xls",sheet_name="DWH")
#选择主键
Primarykey="员工编号"#input("选择主键1:")
Primarykey
# 员工编号

一、数据量

输出表格1–数据量

def write_to_excel_DataVolume(Data,TargetPath): # cor_df 为要保存的 dataframe 
    writer = pd.ExcelWriter(TargetPath, engine='xlsxwriter') # 这里用
    Data.to_excel(writer,sheet_name='Sheet1', encoding='utf8', header=False, startcol=0, startrow=2) # 把dataframe的数据从第2行开始
    workbook  = writer.book
    
    format1 = workbook.add_format({ # 先把样式打包,然后之后赋值即可
        'bold': True, # 字体加粗
        'text_wrap': True, # 是否自动换行
        'valign': 'bottom',  #垂直对齐方式
        'align': 'center', # 水平对齐方式
        'fg_color': '#C5D9F1', # 单元格背景颜色
        'border': 1,# 边框
    })    
    writer_sheet = writer.sheets['Sheet1']
    # 设置宽度
    writer_sheet.set_column("A:I", 16)
    writer_sheet.set_column('C:C',30)
    writer_sheet.merge_range(0,0,0,2,'对比结果',format1)
    writer_sheet.merge_range(4,2,4,0,'数据量差异',format1)
    writer_sheet.write(1,0,'',format1)
    writer_sheet.write(1,1,'ODS',format1)
    writer_sheet.write(1,2,'DWH',format1)
    writer.save()
    writer.close()
DataFrame_DataVolume=pd.DataFrame([[DATA_ODS.shape[0]],[DATA_DWH.shape[0]]]).T
DataFrame_DataVolume.columns =["ODS","DWH"]
DataFrame_DataVolume.index=["数据量"]
DataFrame_DataVolume
#writeFileDataVolume(DataFrame_DataVolume,TargetPath)
write_to_excel_DataVolume(DataFrame_DataVolume,TargetPath)

在这里插入图片描述

输出表格2–数据量差异合同

if DATA_ODS.shape[0]==DATA_DWH.shape[0]:
    pass
else:
    
    DATA_ODS_Primarykey=pd.DataFrame(DATA_ODS[Primarykey])
    DATA_DWH_Primarykey=pd.DataFrame(DATA_DWH[Primarykey])
    df_union = pd.concat([DATA_ODS_Primarykey,DATA_DWH_Primarykey])
    # 实现1
    df_diff_ODS = df_union.append(DATA_ODS_Primarykey).drop_duplicates(subset=df_union.columns.to_list(), keep=False)
    df_diff_DWH = df_union.append(DATA_DWH_Primarykey).drop_duplicates(subset=df_union.columns.to_list(), keep=False)
    #DWH多的合同
    df_diff_ODS
    #DWH少的合同
    df_diff_DWH
    df_diff_DWH_Data=[]
    df_diff_ODS_Data=[]
    for i in df_diff_ODS.head(10).values.tolist():
        for n in i:
            df_diff_ODS_Data.append(n)
            
    for i in df_diff_DWH.head(10).values.tolist():
            df_diff_DWH_Data.append(n)
    while True:
        if len(df_diff_DWH_Data)>len(df_diff_ODS_Data):
            df_diff_ODS_Data.append("-")
        elif len(df_diff_DWH_Data)< len(df_diff_ODS_Data):
            df_diff_DWH_Data.append("-")
        elif len(df_diff_DWH_Data)== len(df_diff_ODS_Data):
            break
    DataFrame_DataVolume_Count_result=pd.DataFrame(df_diff_DWH_Data,df_diff_ODS_Data).reset_index()
    DataFrame_DataVolume_Count_result.columns=['DWH多的合同','DWH少的的合同']
    DataFrame_DataVolume_Count_result=DataFrame_DataVolume_Count_result.reset_index()
    DataFrame_DataVolume_Count_result.columns=['序号','DWH多的合同','DWH少的的合同']
DataFrame_DataVolume_Count_result
from openpyxl import load_workbook
 
def write_to_excel_Count_result(Data,TargetPath):
    df_Old = pd.DataFrame(pd.read_excel(TargetPath)) #读取原数据文件和表 
    writer = pd.ExcelWriter(TargetPath,engine='openpyxl')
    book=load_workbook(TargetPath)
    writer.book = book
    writer.sheets = dict((ws.title, ws) for ws in book.worksheets)
    df_rows = df_Old.shape[0] #获取原数据的行数
    Data.to_excel(writer,startrow=df_rows+1, index=False,startcol=0,header=True)#将数据写入excel中的aa表,从第一个空行开始写
    writer.save()#保存
write_to_excel_Count_result(DataFrame_DataVolume_Count_result,TargetPath)

在这里插入图片描述

您可能感兴趣的文章:

相关文章