optimize shap response latency

backend/api/shap_routes.py

@@ -1,17 +1,31 @@
 from flask import Blueprint, jsonify, request
+import time
 
 from services.shap_service import shap_service
 
 shap_bp = Blueprint('shap', __name__, url_prefix='/api/shap')
 
 
+def log_shap(route, stage, extra=''):
+    message = f'[SHAP] route={route} stage={stage}'
+    if extra:
+        message = f'{message} {extra}'
+    print(message, flush=True)
+
+
 @shap_bp.route('/global', methods=['GET'])
 def get_global_importance():
+    started_at = time.time()
     try:
         model_type = request.args.get('model', 'random_forest')
+        log_shap('/global', 'start', f'model={model_type}')
         result = shap_service.get_global_importance(model_type)
+        elapsed = round(time.time() - started_at, 3)
+        log_shap('/global', 'success', f'model={model_type} elapsed={elapsed}s')
         return jsonify({'code': 200, 'message': 'success', 'data': result})
     except Exception as e:
+        elapsed = round(time.time() - started_at, 3)
+        log_shap('/global', 'error', f'elapsed={elapsed}s error={e}')
         return jsonify({'code': 500, 'message': str(e), 'data': None}), 500
 
 
@@ -30,21 +44,33 @@ def get_local_explanation():
 
 @shap_bp.route('/interaction', methods=['GET'])
 def get_interactions():
+    started_at = time.time()
     try:
         model_type = request.args.get('model', 'random_forest')
         top_n = int(request.args.get('top_n', 10))
+        log_shap('/interaction', 'start', f'model={model_type} top_n={top_n}')
         result = shap_service.get_interactions(model_type, top_n)
+        elapsed = round(time.time() - started_at, 3)
+        log_shap('/interaction', 'success', f'model={model_type} elapsed={elapsed}s')
         return jsonify({'code': 200, 'message': 'success', 'data': result})
     except Exception as e:
+        elapsed = round(time.time() - started_at, 3)
+        log_shap('/interaction', 'error', f'elapsed={elapsed}s error={e}')
         return jsonify({'code': 500, 'message': str(e), 'data': None}), 500
 
 
 @shap_bp.route('/dependence', methods=['GET'])
 def get_dependence():
+    started_at = time.time()
     try:
         feature = request.args.get('feature', '月均加班时长')
         model_type = request.args.get('model', 'random_forest')
+        log_shap('/dependence', 'start', f'model={model_type} feature={feature}')
         result = shap_service.get_dependence(feature, model_type)
+        elapsed = round(time.time() - started_at, 3)
+        log_shap('/dependence', 'success', f'model={model_type} elapsed={elapsed}s')
         return jsonify({'code': 200, 'message': 'success', 'data': result})
     except Exception as e:
+        elapsed = round(time.time() - started_at, 3)
+        log_shap('/dependence', 'error', f'elapsed={elapsed}s error={e}')
         return jsonify({'code': 500, 'message': str(e), 'data': None}), 500

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'}