写excel文件-xlsxwriter包的使用
# encoding: utf8
from xlsxwriter.utility import xl_rowcol_to_cell
import pandas as pd
def df_to_excel(df, writer, sheetname, index=True, header=True, startrow=0, startcol=0):
'''
please refer the method to_excel from module pandas.DataFrame
'''
workbook = writer.book
if sheetname not in writer.sheets:
pd.DataFrame().to_excel(writer, sheetname)
worksheet = writer.sheets[sheetname]
format_title = workbook.add_format({
'border': 1, 'align': 'center', 'bg_color': 'cccccc', 'bold': True})
format_data = workbook.add_format({'border': 1})
title = list(df.columns.get_values())
if index is True:
title = list(df.columns.names) + title
title[0] = title[0] or 'index'
if header is True:
cell = xl_rowcol_to_cell(startrow, startcol) # 'A1'
worksheet.write_row(cell, title, format_title)
startrow += 1
if index is True:
cell = xl_rowcol_to_cell(startrow, startcol) # 'A2'
worksheet.write_column(cell, df.index.get_values(), format_data)
startcol += 1
for i, col in enumerate(df.columns):
cell = xl_rowcol_to_cell(startrow, startcol+i) # 'B2'
worksheet.write_column(cell, df[col], format_data)
if __name__ == "__main__":
X4 = pd.DataFrame({'X1': [1, 2], 'X2': '自'})
with pd.ExcelWriter('f2.xlsx', engine='xlsxwriter') as writer:
# X4.to_excel(writer,sheetname='X4', index=False, header=False, startrow=2, startcol=2)
df_to_excel(X4, writer, sheetname='X4', index=False, header=True, startrow=2, startcol=2)
worksheet = writer.sheets['X4']
worksheet.hide_gridlines(2)
--- 她说, 她是仙,她不是神