2025-03-03 13:21:55 +08:00
|
|
|
|
import json
|
|
|
|
|
|
import os
|
|
|
|
|
|
import random
|
|
|
|
|
|
|
|
|
|
|
|
# 目录路径
|
|
|
|
|
|
directory = "output/uie"
|
|
|
|
|
|
|
|
|
|
|
|
# 确保目录存在
|
|
|
|
|
|
if not os.path.exists(directory):
|
|
|
|
|
|
os.makedirs(directory)
|
|
|
|
|
|
|
|
|
|
|
|
# 读取 JSON 文件
|
|
|
|
|
|
def load_json(file_path):
|
|
|
|
|
|
with open(file_path, 'r', encoding='utf-8') as f:
|
|
|
|
|
|
return json.load(f)
|
|
|
|
|
|
|
2025-03-17 16:10:03 +08:00
|
|
|
|
#互联网,天气查询,知识问答不进行槽位抽取训练
|
|
|
|
|
|
def filter_data(data):
|
|
|
|
|
|
converted_list = []
|
|
|
|
|
|
for item in data:
|
|
|
|
|
|
# if "text" in item and "label" in item:
|
|
|
|
|
|
# converted_list.append({
|
|
|
|
|
|
# "text": item["text"],
|
|
|
|
|
|
# "label": item["label"] # prompt→ label
|
|
|
|
|
|
# })
|
|
|
|
|
|
if "text" in item and "prompt" in item:
|
|
|
|
|
|
converted_list.append(item)
|
|
|
|
|
|
return converted_list
|
|
|
|
|
|
|
2025-03-03 13:21:55 +08:00
|
|
|
|
# 按7:3比例随机拆分 JSON 文件
|
|
|
|
|
|
def split_json_random(input_file, output_file1, output_file2):
|
|
|
|
|
|
# 读取数据
|
|
|
|
|
|
data = load_json(input_file)
|
|
|
|
|
|
|
2025-03-17 16:10:03 +08:00
|
|
|
|
# filter数据
|
|
|
|
|
|
converted_data = filter_data(data)
|
|
|
|
|
|
|
|
|
|
|
|
# 打乱数据顺序
|
|
|
|
|
|
random.shuffle(converted_data)
|
|
|
|
|
|
|
2025-03-03 13:21:55 +08:00
|
|
|
|
# 随机打乱数据
|
2025-03-17 16:10:03 +08:00
|
|
|
|
random.shuffle(converted_data)
|
2025-03-03 13:21:55 +08:00
|
|
|
|
|
|
|
|
|
|
# 计算数据的分割点
|
2025-03-17 16:10:03 +08:00
|
|
|
|
split_point = int(len(converted_data) * 0.7)
|
2025-03-03 13:21:55 +08:00
|
|
|
|
|
|
|
|
|
|
# 按比例分割数据
|
2025-03-17 16:10:03 +08:00
|
|
|
|
data_part1 = converted_data[:split_point] # 70% 训练数据
|
|
|
|
|
|
data_part2 = converted_data[split_point:] # 30% 验证数据
|
2025-03-03 13:21:55 +08:00
|
|
|
|
|
|
|
|
|
|
# 保存数据到两个文件
|
|
|
|
|
|
with open(output_file1, 'w', encoding='utf-8') as f1:
|
|
|
|
|
|
json.dump(data_part1, f1, ensure_ascii=False, indent=4)
|
|
|
|
|
|
|
|
|
|
|
|
with open(output_file2, 'w', encoding='utf-8') as f2:
|
|
|
|
|
|
json.dump(data_part2, f2, ensure_ascii=False, indent=4)
|
|
|
|
|
|
|
2025-03-17 16:10:03 +08:00
|
|
|
|
print(f"数据已随机打乱并按 7:3 分割,保存至:\n - {output_file1}({len(data_part1)} 条)\n - {output_file2}({len(data_part2)} 条)")
|
2025-03-03 13:21:55 +08:00
|
|
|
|
|
|
|
|
|
|
# 输入的 JSON 文件路径
|
|
|
|
|
|
input_file = 'output/merged_data.json'
|
|
|
|
|
|
# 输出的两个文件路径
|
|
|
|
|
|
output_file1 = 'output/uie/train.json'
|
|
|
|
|
|
output_file2 = 'output/uie/val.json'
|
|
|
|
|
|
|
|
|
|
|
|
# 按 7:3 随机拆分并保存
|
|
|
|
|
|
split_json_random(input_file, output_file1, output_file2)
|