import pandas as pd import os def convert_file_to_json(file_path): # 检查文件扩展名并读取文件 _, file_extension = os.path.splitext(file_path) if file_extension.lower() == '.csv': df = pd.read_csv(file_path) elif file_extension.lower() in ['.xlsx', '.xls']: df = pd.read_excel(file_path) else: raise ValueError("Unsupported file format. Only CSV and XLSX files are supported.") # 将\N、NaN值、空字符串、换行符等转换为空字符串 df = df.replace(r'\N', '', regex=False) # 替换\N为空字符串 df = df.fillna('') # 替换NaN值为空字符串 df = df.replace('\n', '', regex=True) # 替换换行符为空字符串 df = df.replace('undefined', '', regex=True) # 替换换行符为空字符串 # 将DataFrame转换为JSON格式的字符串 json_str = df.to_json(orient='records', lines=False, force_ascii=False) # 保存为JSON文件,文件名与原文件相同,但扩展名为.json json_file_path = os.path.splitext(file_path)[0] + '.json' with open(json_file_path, 'w', encoding='utf-8') as json_file: json_file.write(json_str) print(f"File {file_path} has been successfully converted to JSON and saved as {json_file_path}") # 使用示例 file_path = '装修.xlsx' convert_file_to_json(file_path)
标签:为空,df,MongoDB,Excel,后缀名,json,file,字符串,path From: https://www.cnblogs.com/wolvies/p/18544092