python openpyxl 操作 excel

it2024-10-04  19

初识与安装

Openpyxl is a Python library for reading and writing Excel 2010 xlsx/xlsm/xltx/xltm files.

安装

$ pip install openpyxl

  

一个简单创建例子

from openpyxl import Workbook wb = Workbook() # 激活 worksheet ws = wb.active # 数据可以直接分配到单元格中 ws['A1'] = 42 # 可以附加行,从第一列开始附加 ws.append([1, 2, 3]) # Python 类型会被自动转换 import datetime ws['A3'] = datetime.datetime.now().strftime("%Y-%m-%d") # 保存文件 wb.save("sample.xlsx")

打开查看Excel如下:

  

workbook

There is no need to create a file on the filesystem to get started with openpyxl. Just import the Workbook class and start using it.

from openpyxl import Workbook wb = Workbook()

  

worksheet

A workbook至少创建一个worksheet.

通过openpyxl.workbook.Workbook.active()得到worksheet.ws = wb.active注意:该方法使用_active_sheet_index属性, 默认会设置0,也就是第一个worksheet。除非手动修改,否则使用active方法得到都是第一个worksheet。你也可以创建worksheets,通过 openpyxl.workbook.Workbook.create_sheet() 方法:

>>> ws1 = wb.create_sheet("Mysheet") #插入到最后(default) #或者 >>> ws2 = wb.create_sheet("Mysheet", 0) #插入到最开始的位置

创建的sheet的名称会自动创建,按照sheet,sheet1,sheet2自动增长,通过title属性可以修改其名称。ws.title = "New Title"默认的sheet的tab是白色的,可以通过 RRGGBB颜色来修改sheet_properties.tabColor属性从而修改sheet tab按钮的颜色:ws.sheet_properties.tabColor = "1072BA"当你设置了sheet的名称,可以将其看成workbook中的一个key。也可以使用openpyxl.workbook.Workbook.get_sheet_by_name() 方法

 

>>> ws3 = wb["New Title"] >>> ws4 = wb.get_sheet_by_name("New Title") >>> ws is ws3 is ws4 True

查看workbook中的所有worksheets名称:openpyxl.workbook.Workbook.get_sheet_names()

>>> print(wb.sheetnames) ['Sheet2', 'New Title', 'Sheet1']

遍历worksheets:

>>> for sheet in wb: ... print(sheet.title)

  

操作数据

访问单元格

单元格可以看作是worksheet的key,通过key去访问单元格中的数据

>>> c = ws['A4']

直接返回A4单元格,如果不存在则会自动创建一个。

指定单元格的值

>>> ws['A4'] = 4 #直接赋值

使用openpyxl.worksheet.Worksheet.cell()方法操作某行某列的某个值:

>>> d = ws.cell(row=4, column=2, value=10)

注意:

当worksheet在内存中被创建时,是没有包含cells的,cells是在首次访问时创建.可以循环在内存中创建cells,这时不指定他们的值也会创建该cells些:(创建100x100cells) >>> for i in range(1,101): ... for j in range(1,101): ... ws.cell(row=i, column=j)

访问许多cells

通过切片Ranges指定许多cells

>>> cell_range = ws['A1':'C2']

同样也可以Ranges rows 或者columns :

>>> colC = ws['C'] >>> col_range = ws['C:D'] >>> row10 = ws[10] >>> row_range = ws[5:10]

也可以使用 openpyxl.worksheet.Worksheet.iter_rows() 方法:(需要指定行->行,截止列)

>>> for row in ws.iter_rows(min_row=1, max_col=3, max_row=2): ... for cell in row: ... print(cell) <Cell Sheet1.A1> <Cell Sheet1.B1> <Cell Sheet1.C1> <Cell Sheet1.A2> <Cell Sheet1.B2> <Cell Sheet1.C2>  也可以使用  openpyxl.worksheet.Worksheet.iter_cols() 方法:(需要指定列->列,截止行) >>> for row in ws.iter_rows(min_row=1, max_col=3, max_row=2): ... for cell in row: ... print(cell) <Cell Sheet1.A1> <Cell Sheet1.B1> <Cell Sheet1.C1> <Cell Sheet1.A2> <Cell Sheet1.B2> <Cell Sheet1.C2>

如果你需要遍历所有文件的行或列,可以使用openpyxl.worksheet.Worksheet.rows() 属性:

>>> ws = wb.active >>> ws['C9'] = 'hello world' >>> tuple(ws.rows) ((<Cell Sheet.A1>, <Cell Sheet.B1>, <Cell Sheet.C1>), (<Cell Sheet.A2>, <Cell Sheet.B2>, <Cell Sheet.C2>), (<Cell Sheet.A3>, <Cell Sheet.B3>, <Cell Sheet.C3>), (<Cell Sheet.A4>, <Cell Sheet.B4>, <Cell Sheet.C4>), (<Cell Sheet.A5>, <Cell Sheet.B5>, <Cell Sheet.C5>), (<Cell Sheet.A6>, <Cell Sheet.B6>, <Cell Sheet.C6>), (<Cell Sheet.A7>, <Cell Sheet.B7>, <Cell Sheet.C7>), (<Cell Sheet.A8>, <Cell Sheet.B8>, <Cell Sheet.C8>), (<Cell Sheet.A9>, <Cell Sheet.B9>, <Cell Sheet.C9>))

或者 openpyxl.worksheet.Worksheet.columns() 属性:

>>> tuple(ws.columns) ((<Cell Sheet.A1>, <Cell Sheet.A2>, <Cell Sheet.A3>, <Cell Sheet.A4>, <Cell Sheet.A5>, <Cell Sheet.A6>, ... <Cell Sheet.B7>, <Cell Sheet.B8>, <Cell Sheet.B9>), (<Cell Sheet.C1>, <Cell Sheet.C2>, <Cell Sheet.C3>, <Cell Sheet.C4>, <Cell Sheet.C5>, <Cell Sheet.C6>, <Cell Sheet.C7>, <Cell Sheet.C8>, <Cell Sheet.C9>))

Saving to a file

最简单最安全的方法保存workbook是使用openpyxl.workbook.Workbook对象的 openpyxl.workbook.Workbook.save()方法:

>>> wb = Workbook() >>> wb.save('balances.xlsx')

保存的默认位置在python的根目录下。注意:会自动覆盖已经存在文件名的文件。

Loading from a file

像写一样我们可以导入openpyxl.load_workbook()已经存在的workbook:

>>> from openpyxl import load_workbook >>> wb2 = load_workbook('test.xlsx') >>> print wb2.get_sheet_names() ['Sheet2', 'New Title', 'Sheet1']

  

常用实例

详情参考官方使用文档:

http://openpyxl.readthedocs.io/en/default/usage.html

写入例子一

#!/usr/bin/env python # -*- coding: utf-8 -*- from openpyxl import Workbook wb = Workbook() # 激活 worksheet ws = wb.active # 数据可以直接分配到单元格中 ws['A1'] = 42 # 可以附加行,从第一列开始附加 ws.append([1, 2, 3]) # Python 类型会被自动转换 import datetime ws['A3'] = datetime.datetime.now().strftime("%Y-%m-%d") # 保存文件 wb.save("sample.xlsx")

写入例子二

#!/usr/bin/env python # -*- coding: utf-8 -*- """ http://openpyxl.readthedocs.io/en/default/usage.html """ # workbook相关 from openpyxl import Workbook from openpyxl.compat import range from openpyxl.utils import get_column_letter wb = Workbook() dest_filename = 'empty_book.xlsx' ws1 = wb.active ws1.title = "range names" for row in range(1, 40): ws1.append(range(600)) ws2 = wb.create_sheet(title="Pi") ws2['F5'] = 3.14 ws3 = wb.create_sheet(title="Data") for row in range(10, 20): for col in range(27, 54): _ = ws3.cell(column=col, row=row, value="{0}".format(get_column_letter(col))) print(ws3['AA10'].value) wb.save(filename=dest_filename)

读取例子一

#!/usr/bin/env python # -*- coding: utf-8 -*-from openpyxl.reader.excel import load_workbookimport json# 读取excel2007文件wb = load_workbook(filename=r'test_book.xlsx')# 显示有多少张表print "Worksheet range(s):", wb.get_named_ranges()print "Worksheet name(s):", wb.get_sheet_names()# 取第一张表sheetnames = wb.get_sheet_names()ws = wb.get_sheet_by_name(sheetnames[0])# 显示表名,表行数,表列数print "Work Sheet Titile:", ws.titleprint "Work Sheet Rows:", ws.max_rowprint "Work Sheet Cols:", ws.max_column# 建立存储数据的字典data_dic = {}# 把数据存到字典中for rx in range(1, ws.max_row + 1): temp_list = [] pid = rx w1 = ws.cell(row=rx, column=1).value w2 = ws.cell(row=rx, column=2).value w3 = ws.cell(row=rx, column=3).value w4 = ws.cell(row=rx, column=4).value temp_list = [w1, w2, w3, w4] data_dic[pid] = temp_list# 打印字典数据个数print 'Total:%d' % len(data_dic)print json.dumps(data_dic, encoding="UTF-8", ensure_ascii=False)

 读取结果:

Worksheet range(s): [] Worksheet name(s): [u'\u6d3b\u52a8\u8868', u'\u7528\u6237\u4fe1\u606f', u'Sheet3'] Work Sheet Titile: 活动表 Work Sheet Rows: 3 Work Sheet Cols: 5 Total:3 {"1": ["张三", 18, "男", "广州"], "2": ["李四", 20, "女", "湖北"], "3": ["王五", 25, "女", "北京"]}

  

使用公式

>>> from openpyxl import Workbook >>> wb = Workbook() >>> ws = wb.active >>> # add a simple formula >>> ws["A1"] = "=SUM(1, 1)" >>> wb.save("formula.xlsx")

 

 

 


 

***微信扫一扫,关注“python测试开发圈”,了解更多测试教程!***

转载于:https://www.cnblogs.com/guanfuchang/p/5970435.html

最新回复(0)