调优，差训练过程图

api_send_evaluate.py

@@ -18,9 +18,9 @@ def download_file(data_file_info, save_dir):
         log_file_url = data_file_info["log_file_url"]
         log_save_path = os.path.join(save_dir, "log.txt")
         response = requests.get(log_file_url)
-        with open(log_save_path, 'wb') as file:
+        # with open(log_save_path, 'wb') as file:
-            file.write(response.content)
+        #     file.write(response.content)
-        print(f"Log file saved to: {log_save_path}")
+        # print(f"Log file saved to: {log_save_path}")
     else:
         print("无法获取日志文件信息")
 
@@ -29,9 +29,9 @@ def download_file(data_file_info, save_dir):
         data_file_url = data_file_info["data_file_url"]
         data_save_path = os.path.join(save_dir, "data.xlsx")
         response = requests.get(data_file_url)
-        with open(data_save_path, 'wb') as file:
+        # with open(data_save_path, 'wb') as file:
-            file.write(response.content)
+        #     file.write(response.content)
-        print(f"Data file saved to: {data_save_path}")
+        # print(f"Data file saved to: {data_save_path}")
     else:
         print("无法获取数据文件信息")
 

api_send_train.py

@@ -16,9 +16,9 @@ def download_file(data_file_info, save_dir):
         model_file_url = data_file_info["model_file_url"]
         model_save_path = os.path.join(save_dir, "model.pth")
         response = requests.get(model_file_url)
-        with open(model_save_path, 'wb') as file:
+        # with open(model_save_path, 'wb') as file:
-            file.write(response.content)
+        #     file.write(response.content)
-        print(f"Model saved to: {model_save_path}")
+        # print(f"Model saved to: {model_save_path}")
     else:
         print("无法获取模型文件信息")
 
@@ -27,9 +27,9 @@ def download_file(data_file_info, save_dir):
         log_file_url = data_file_info["log_file_url"]
         log_save_path = os.path.join(save_dir, "log.txt")
         response = requests.get(log_file_url)
-        with open(log_save_path, 'wb') as file:
+        # with open(log_save_path, 'wb') as file:
-            file.write(response.content)
+        #     file.write(response.content)
-        print(f"Log file saved to: {log_save_path}")
+        # print(f"Log file saved to: {log_save_path}")
     else:
         print("无法获取日志文件信息")
 
@@ -38,9 +38,9 @@ def download_file(data_file_info, save_dir):
         data_file_url = data_file_info["data_file_url"]
         data_save_path = os.path.join(save_dir, "data.xlsx")
         response = requests.get(data_file_url)
-        with open(data_save_path, 'wb') as file:
+        # with open(data_save_path, 'wb') as file:
-            file.write(response.content)
+        #     file.write(response.content)
-        print(f"Data file saved to: {data_save_path}")
+        # print(f"Data file saved to: {data_save_path}")
     else:
         print("无法获取数据文件信息")
 

config/config.yaml

@@ -23,21 +23,19 @@ paths:
 
 #---训练配置---#
 training:
-  n_epochs: 300
+  n_epochs: 500
-  batch_size: 16 # CPU
+  batch_size: 1024
-  learning_rate: 0.001 # CPU
+  learning_rate: 0.001
-  # batch_size: 128 # GPU
+  early_stop_patience: 100
-  # learning_rate: 0.001 # GPU
-  early_stop_patience: 50
   scheduler:
     gamma: 0.98
-    step_size: 10
+    step_size: 20
   regularization:
-    l1_lambda: 1e-5
+    l1_lambda: 1e-6 # 默认1e-5, 降低L1，因为特征已经通过权重控制
-    l2_lambda: 1e-4
+    l2_lambda: 5e-4 # 默认1e-4, 增加L2，加强过拟合控制
   dropout_rate: 0.2
-  experimental_mode: false
+  experimental_mode: true
-  experiments_count: 50
+  experiments_count: 10
   replace_model: true
   data_mode: "train_val"  # 可选: "train", "train_val", "all"
 
@@ -45,7 +43,7 @@ training:
 model:
   num_classes: 4  # nc
   input_dim: 10
-  architecture: "mlp"  # 可选: "mlp", "transformer"
+  architecture: "transformer"  # 可选: "mlp", "transformer"
   mlp:
     layers:
       - output_dim: 32

utils/model_trainer.py

@@ -324,7 +324,17 @@ class MLModel:
         torch.save(self.model.state_dict(), path)
 
     def train_detect(self) -> Tuple[float, float, List[float], List[float], List[float]]:
-        """训练和检测模型"""
+        """
+        训练和检测模型
+        
+        Returns:
+            Tuple[float, float, List[float], List[float], List[float]]:
+                - average_f1: 平均F1分数
+                - wrong_percentage: 错误率
+                - precision_scores: 各类别精确率
+                - recall_scores: 各类别召回率
+                - f1_scores: 各类别F1分数
+        """
         # 从配置文件中读取数据路径
         data = pd.read_excel(self.config['data_path'])
         
@@ -377,7 +387,9 @@ class MLModel:
         if self.config['training']['experimental_mode']:
             return self._run_experiments(X_train, y_train, X_val, y_val, class_weights)
         else:
-            return self._single_train_detect(X_train, y_train, X_val, y_val, class_weights)
+            results = self._single_train_detect(X_train, y_train, X_val, y_val, class_weights)
+            # 只返回前5个值，去掉 best_epoch
+            return results[:-1]
 
     def _update_metrics(self, train_metrics: Dict, val_metrics: Dict,
                        train_loss: float, train_acc: float,
@@ -476,7 +488,11 @@ class MLModel:
         logging.info(log_message)
         print(log_message)
         
-        return best_experiment_results[:-1]
+        return (best_experiment_results[0],  # average_f1
+                best_experiment_results[1],  # wrong_percentage
+                best_experiment_results[2],  # precision
+                best_experiment_results[3],  # recall
+                best_experiment_results[4])  # f1
 
     def _single_train_detect(self, X_train, y_train, X_val, y_val, class_weights, model_path=None):
         """单次训练过程"""