⚡ Bolt: LRUキャッシュによるML推論の最適化とNumPyオーバーヘッドの削減

.jules/bolt.md

@@ -9,3 +9,7 @@
 ## 2026-06-10 - [Lazy Loading Scikit-learn & NumPy]
 **Learning:** Top-level imports of `numpy` and `joblib` plus loading a serialized model file (`.joblib`) during FastAPI startup adds significant overhead (e.g., ~3 seconds). Refactoring this into a service with `cached_property` and local imports deferred the cost until the first request.
 **Action:** Move all heavy ML model loading and their dependencies into lazy-loaded properties to ensure near-instant application startup.
+
+## 2026-06-11 - [ML Inference Caching & NumPy Overhead]
+**Learning:** Using a per-instance `lru_cache` (via a `cached_property` returning a decorated inner function) for ML inference avoids "unhashable self" errors and prevents cross-instance memory leaks. Additionally, for single-sample predictions in Scikit-learn, passing a list of tuples directly is faster than the overhead of NumPy array allocation and reshaping.
+**Action:** Implement per-instance caching for model inference and evaluate if passing raw Python types to ML models is more efficient than NumPy for small/single-sample inputs.

ml_service.py

@@ -1,4 +1,4 @@
-from functools import cached_property
+from functools import cached_property, lru_cache
 import logging
 
 logger = logging.getLogger(__name__)
@@ -15,12 +15,22 @@ def model(self):
             logger.error(f"Error loading ML model: {e}")
             return None
 
+    @cached_property
+    def _cached_predict(self):
+        """Internal cached prediction function to avoid unhashable 'self'."""
+        @lru_cache(maxsize=128)
+        def _predict(features_tuple):
+            # Pass a list containing the features tuple directly to the model.
+            # This avoids the overhead of numpy array creation and reshaping.
+            prediction = self.model.predict([features_tuple])
+            return int(prediction[0])
+        return _predict
+
     def predict(self, features_list):
-        """Perform prediction using the lazy-loaded model."""
+        """Perform prediction using the lazy-loaded model with caching."""
         if self.model is None:
             raise RuntimeError("ML model could not be loaded")
 
-        import numpy as np
-        features = np.array(features_list).reshape(1, -1)
-        prediction = self.model.predict(features)
-        return int(prediction[0])
+        # Convert list to tuple to make it hashable for LRU cache
+        features_tuple = tuple(features_list)
+        return self._cached_predict(features_tuple)