optimize shap response latency

backend/core/shap_analysis.py

@@ -23,7 +23,8 @@ class SHAPAnalyzer:
         self.feature_names = None
         self.selected_features = None
         self.label_encoders = {}
-        self.background_data = None
+        self.background_data = {}
+        self.global_result_cache = {}
         self._initialized = False
 
     def _ensure_initialized(self):
@@ -85,8 +86,8 @@ class SHAPAnalyzer:
 
     def _get_background_sample(self, n_samples=500):
         """获取背景数据样本"""
-        if self.background_data is not None:
-            return self.background_data
+        if n_samples in self.background_data:
+            return self.background_data[n_samples]
 
         try:
             from core.preprocessing import get_clean_data
@@ -123,7 +124,7 @@ class SHAPAnalyzer:
                 if selected_indices:
                     X = X[:, selected_indices]
 
-            self.background_data = X
+            self.background_data[n_samples] = X
             return X
         except Exception:
             return None
@@ -151,12 +152,15 @@ class SHAPAnalyzer:
         if not SHAP_AVAILABLE:
             return {'error': 'SHAP library not installed'}
 
+        if model_type in self.global_result_cache:
+            return self.global_result_cache[model_type]
+
         self._ensure_initialized()
         explainer = self._get_tree_explainer(model_type)
         if explainer is None:
             return {'error': f'No tree model available for {model_type}'}
 
-        X = self._get_background_sample()
+        X = self._get_background_sample(n_samples=32)
         if X is None:
             return {'error': 'Failed to prepare background data'}
 
@@ -215,11 +219,13 @@ class SHAPAnalyzer:
                     'dimension': self._map_feature_to_dimension(fname),
                 })
 
-            return {
+            result = {
                 'model_type': model_type,
                 'dimensions': dimensions,
                 'top_features': top_features,
             }
+            self.global_result_cache[model_type] = result
+            return result
         except Exception as exc:
             return {'error': str(exc)}
 
@@ -306,7 +312,7 @@ class SHAPAnalyzer:
         if explainer is None:
             return {'error': f'No tree model available for {model_type}'}
 
-        X = self._get_background_sample(n_samples=200)
+        X = self._get_background_sample(n_samples=12)
         if X is None:
             return {'error': 'Failed to prepare background data'}
 
@@ -362,7 +368,7 @@ class SHAPAnalyzer:
         if explainer is None:
             return {'error': f'No tree model available for {model_type}'}
 
-        X = self._get_background_sample()
+        X = self._get_background_sample(n_samples=24)
         if X is None:
             return {'error': 'Failed to prepare background data'}