Merge pull request #26 from pekopoke/dev

Dev:add gt and pre of code formula table text in result jsonl

Author: e06084

examples/basic_usage.py

@@ -804,7 +804,7 @@ def demo_multi_extraction():
     # 配置文件路径
     data_dir = Path("../data")
     # dataset_path = data_dir / "sample_dataset.jsonl"
-    dataset_path = "/home/lulindong/Pycharm_projects/cc/WebMainBench_llm-webkit_v1_WebMainBench_dataset_merge_with_llm_webkit.jsonl"
+    dataset_path = "/home/lulindong/Pycharm_projects/cc/WebMainBench_1904_v1_WebMainBench_dataset_merge_with_llm_webkit.jsonl"
 
     print(f"📂 数据集文件: {dataset_path}")
 
@@ -957,7 +957,7 @@ def demo_llm_webkit_with_preprocessed_html_evaluation():
     print("1. 从真实数据集加载预处理HTML数据...")
 
     # 使用DataLoader加载真实的样本数据
-    dataset_path = Path("data/WebMainBench_dataset_sample2.jsonl")
+    dataset_path = Path("/home/lulindong/Pycharm_projects/cc/WebMainBench_1904_v1_WebMainBench_dataset_merge_with_llm_webkit.jsonl")
     print(f"📂 数据集文件: {dataset_path}")
 
     if not dataset_path.exists():
@@ -969,7 +969,6 @@ def demo_llm_webkit_with_preprocessed_html_evaluation():
     dataset = DataLoader.load_jsonl(dataset_path, include_results=False)
     dataset.name = "real_preprocessed_html_test"
     dataset.description = "基于真实数据的预处理HTML功能测试"
-
 
     print(f"✅ 真实数据集加载成功，包含 {len(dataset)} 个样本")
     print("📋 真实数据样本包含:")
@@ -1069,15 +1068,22 @@ def demo_llm_webkit_with_preprocessed_html_evaluation():
             print(f"  ⏱️  提取时间: {sample_result.get('extraction_time', 0):.3f}秒")
         else:
             print(f"  ❌ 提取失败")
-    
     # 7. 保存结果
     print(f"\n7. 💾 保存评测结果...")
 
     results_dir = Path("results")
     results_dir.mkdir(exist_ok=True)
-    
-    results_path = results_dir / "preprocessed_html_evaluation_results.json"
-    report_path = results_dir / "preprocessed_html_evaluation_report.csv"
+    # 新增：保存带抽取结果的增强数据集（JSONL格式）
+    jsonl_dataset_path = results_dir / f"{extractor.name}_preprocessed_html_dataset_with_results.jsonl"
+    DataSaver.save_dataset_with_extraction(
+        results=result,
+        dataset=dataset,  # 原始数据集对象
+        file_path=jsonl_dataset_path,
+        extractor_name="llm-webkit"  # 抽取器名称前缀
+    )
+    print(f"✅ 带抽取结果的JSONL数据集已保存到: {jsonl_dataset_path}")
+    results_path = results_dir / f"{extractor.name}_preprocessed_html_evaluation_results.json"
+    report_path = results_dir / f"{extractor.name}_preprocessed_html_evaluation_report.csv"
 
     DataSaver.save_evaluation_results(result, results_path)
     DataSaver.save_summary_report(result, report_path)

results/preprocessed_html_evaluation_report.csv

@@ -1,2 +1,2 @@
-extractor,dataset,total_samples,success_rate,overall,code_edit,formula_edit,table_TEDS,table_edit,text_edit
-llm-webkit,real_preprocessed_html_test,2,1.0,0.0091,0.0,0.0,0.0,0.0,0.0453
+extractor,dataset,total_samples,success_rate,overall,code_edit,formula_edit,table_TEDS,table_edit,text_edit
+llm-webkit,real_preprocessed_html_test,2,1.0,0.0091,0.0,0.0,0.0,0.0,0.0453

webmainbench/data/saver.py

@@ -239,6 +239,7 @@ def save_dataset_with_extraction(results: Union["EvaluationResult", Dict[str, An
 
             # Add extraction results if available
             extraction_result = extraction_map.get(sample.id)
+            from webmainbench.metrics.base import BaseMetric
             if extraction_result:
                 # Add extracted content with extractor name prefix
                 sample_dict[f'{extractor_name}_content'] = extraction_result.get('extracted_content', '')
@@ -251,7 +252,22 @@ def save_dataset_with_extraction(results: Union["EvaluationResult", Dict[str, An
                 for metric_name, metric_data in metrics.items():
                     if isinstance(metric_data, dict) and metric_data.get('success', False):
                         sample_dict[f'{extractor_name}_{metric_name}_score'] = metric_data.get('score', 0)
-            
+
+                # 解析预测值（predicted）
+                predicted_content = extraction_result.get('extracted_content', '')
+                predicted_parts = BaseMetric._extract_from_markdown(predicted_content)  # 关键：解析预测内容
+                for part_type in ['code', 'formula', 'table', 'text']:
+                    sample_dict[f'{extractor_name}_predicted_{part_type}'] = predicted_parts.get(part_type, '')
+
+                # 解析真实值（groundtruth）
+                groundtruth_content = sample_dict.get('groundtruth_content', '')
+                groundtruth_parts = BaseMetric._extract_from_markdown(groundtruth_content)  # 关键：解析真实内容
+                for part_type in ['code', 'formula', 'table', 'text']:
+                    sample_dict[f'{extractor_name}_groundtruth_{part_type}'] = groundtruth_parts.get(part_type,
+                                                                                                     '')
+
+
+
             enriched_samples.append(sample_dict)
 
         # Save as JSONL