ouyangyouzhang/QueryRewrite
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

优化意图识别示例，更新文档相关性判断逻辑，增强Excel数据验证功能，改进日志记录，调整参数以提升代码可读性和灵活性。

Browse Source
This commit is contained in:
ouyangyouzhang
2025-06-25 09:10:28 +08:00
parent 7142c7c43e
commit 33bc91f0fe
4 changed files with 35 additions and 55 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rag2_0/demo/validate_excel_data_batch.py
+20 -33
View File
@@ -33,7 +33,7 @@ class ValidationResult(BaseModel):
class ExcelDataValidator:
"""Excel数据验证类，用于批量验证Excel数据中的问题分类、问题拆解、检索关键词和问题改写"""
def __init__(self, input_file=None, output_file=None, workers=4, batch_size=10, debug=False):
def __init__(self, input_file=None, output_file=None, workers=4, debug=False):
"""
初始化验证器
@@ -41,7 +41,6 @@ class ExcelDataValidator:
input_file: 输入Excel文件路径
output_file: 输出结果Excel文件路径
workers: 并行工作线程数
batch_size: 每批处理的行数
debug: 是否启用调试模式（串行处理）
"""
# 加载环境变量
@@ -50,7 +49,6 @@ class ExcelDataValidator:
self.input_file = input_file
self.output_file = output_file
self.workers = workers
self.batch_size = batch_size
self.debug = debug
self.df = None
@@ -86,7 +84,7 @@ class ExcelDataValidator:
try:
df = pd.read_excel(file_path)
required_columns = ["问题", "问题分类", "问题改写", "槽点信息"]
required_columns = ["问题", "问题分类", "问题改写", "槽位信息", "检索的内容"]
for col in required_columns:
if col not in df.columns:
logging.error(f"缺少必要的列: {col}", exc_info=True)
@@ -320,7 +318,7 @@ class ExcelDataValidator:
query = row["问题"]
query_class = row.get("问题分类", "")
rewrite = row.get("问题改写", "")
slot_info = row.get("槽点信息", "")
slot_info = row.get("槽位信息", "")
retrieve_content = row.get("检索的内容", "")
if self.debug:
@@ -359,15 +357,16 @@ class ExcelDataValidator:
if len(query_class_list) >= 2:
result = self.validate_classification(llm, rewrite, query_class_list[0], query_class_list[1])
if isinstance(result, tuple) and len(result) >= 3:
is_correct, error_reason, confidence_score = result[:3]
is_correct, error_reason, classification_confidence = result[:3]
confidence_score = max(confidence_score, classification_confidence)
if self.debug:
logging.info(f" 问题分类验证结果: {'通过' if is_correct else '不通过'}, 置信度: {confidence_score:.2f}")
logging.info(f" 问题分类验证结果: {'通过' if is_correct else '不通过'}, 置信度: {classification_confidence:.2f}")
if not is_correct:
logging.info(f" 错误原因: {error_reason}")
if not is_correct:
return index, False, "问题分类", error_reason, confidence_score
return index, False, "问题分类", error_reason, classification_confidence
@@ -416,13 +415,6 @@ class ExcelDataValidator:
logging.error(error_msg, exc_info=True)
return index, False, "处理错误", error_msg, 0.0
def process_batch(self, llm, batch_data):
"""处理一批数据"""
results = []
for row_data in batch_data:
results.append(self.validate_row(llm, row_data))
return results
def create_llm_instances(self, count):
"""创建多个LLM实例"""
api_key = os.getenv("OPENAI_API_KEY")
@@ -437,7 +429,7 @@ class ExcelDataValidator:
return [OpenAiLLM(**llm_params) for _ in range(count)]
def validate(self, input_file=None, output_file=None, workers=None, batch_size=None, debug=None):
def validate(self, input_file=None, output_file=None, workers=None, debug=None):
"""
执行验证过程
@@ -445,7 +437,7 @@ class ExcelDataValidator:
input_file: 输入Excel文件路径
output_file: 输出结果Excel文件路径
workers: 并行工作线程数
batch_size: 每批处理的行数
batch_size: 每批处理的行数（已弃用，保留参数保持兼容）
debug: 是否启用调试模式（串行处理）
Returns:
@@ -454,7 +446,6 @@ class ExcelDataValidator:
input_file = input_file or self.input_file
output_file = output_file or self.output_file
workers = workers or self.workers
batch_size = batch_size or self.batch_size
debug = debug if debug is not None else self.debug
# 读取数据
@@ -492,21 +483,20 @@ class ExcelDataValidator:
# 输出当前结果
logging.info(f"行 {index} 验证结果: {'通过' if is_correct else '不通过'}, 错误环节: {error_phase}, 错误原因: {error_reason}, 置信度: {confidence_score:.2f}")
else:
# 正常模式：并行处理
batches = [all_rows[i:i+batch_size] for i in range(0, len(all_rows), batch_size)]
llm_instances = self.create_llm_instances(min(workers, len(batches)))
# 正常模式：并行处理，每行单独处理
llm_instances = self.create_llm_instances(min(workers, len(all_rows)))
with concurrent.futures.ThreadPoolExecutor(max_workers=workers) as executor:
# 为每个批次分配一个LLM实例
future_to_batch = {
executor.submit(self.process_batch, llm_instances[i % len(llm_instances)], batch):
i for i, batch in enumerate(batches)
# 为每行分配一个LLM实例
future_to_row = {
executor.submit(self.validate_row, llm_instances[i % len(llm_instances)], row_data):
i for i, row_data in enumerate(all_rows)
}
# 使用tqdm显示进度条
for future in tqdm(concurrent.futures.as_completed(future_to_batch), total=len(batches), desc="批次处理进度"):
batch_results = future.result()
all_results.extend(batch_results)
for future in tqdm(concurrent.futures.as_completed(future_to_row), total=len(all_rows), desc="处理进度"):
result = future.result()
all_results.append(result)
# 按行索引排序结果，确保与原始数据顺序一致
all_results.sort(key=lambda x: x[0])
@@ -558,16 +548,14 @@ class ExcelDataValidator:
def main():
"""主函数"""
# 解析命令行参数
input_excel = os.path.join(os.path.dirname(__file__), "..", "..", "data", "excel", "1500条点踩软件问题测试_检索结果.xlsx")
input_excel = os.path.join(os.path.dirname(__file__), "..", "..", "data", "excel", "1500条点踩软件问题测试_意图分类.xlsx")
output_excel = os.path.join(os.path.dirname(__file__), "..", "..", "data", "excel", "自动验证_问题分类重写结果.xlsx")
parser = argparse.ArgumentParser(description="验证Excel数据中的问题分类、问题拆解、检索关键词和问题改写")
parser.add_argument("--input", "-i", type=str, help="输入Excel文件路径", default=input_excel)
parser.add_argument("--output", "-o", type=str, help="输出结果Excel文件路径", default=output_excel)
parser.add_argument("--workers", "-w", type=int, default=20, help="并行工作线程数")
parser.add_argument("--batch-size", "-b", type=int, default=5, help="每批处理的行数")
parser.add_argument("--debug", "-d", action="store_true", help="启用调试模式（串行处理）")
logging.info(f"输入文件路径: {args.input}, 输出文件路径: {args.output}, 并行工作线程数: {args.workers}")
args = parser.parse_args()
is_debug = hasattr(sys, 'gettrace') and sys.gettrace() is not None
@@ -576,7 +564,6 @@ def main():
input_file=args.input,
output_file=args.output,
workers=args.workers,
batch_size=args.batch_size,
debug=is_debug
)
validator.validate()