matrice-analytics 0.1.54__tar.gz → 0.1.55__tar.gz

{matrice_analytics-0.1.54 → matrice_analytics-0.1.55}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: matrice_analytics
-Version: 0.1.54
+Version: 0.1.55
 Summary: Common server utilities for Matrice.ai services
 Author-email: "Matrice.ai" <dipendra@matrice.ai>
 License-Expression: MIT

{matrice_analytics-0.1.54 → matrice_analytics-0.1.55}/matrice_analytics.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: matrice_analytics
-Version: 0.1.54
+Version: 0.1.55
 Summary: Common server utilities for Matrice.ai services
 Author-email: "Matrice.ai" <dipendra@matrice.ai>
 License-Expression: MIT

{matrice_analytics-0.1.54 → matrice_analytics-0.1.55}/src/matrice_analytics/post_processing/face_reg/embedding_manager.py

@@ -205,23 +205,58 @@ class EmbeddingManager:
             
             self.staff_embeddings = []
             embeddings_list = []
-            
+            expected_dim: Optional[int] = None
+            dims_observed: List[int] = []
+            mismatch_examples: List[Tuple[str, int]] = []  # (staffId, dim)
+
             for item in embeddings_data:
-                # if not item.get("isActive", True): # TODO: Check what is isActive
-                #     continue
-                    
+                # Skip inactive if provided
+                if isinstance(item, dict) and item.get("isActive") is False:
+                    continue
+
+                raw_emb = []
+                try:
+                    raw_emb = item.get("embedding", []) if isinstance(item, dict) else []
+                except Exception:
+                    raw_emb = []
+                # Record observed dimension for debugging
+                try:
+                    dims_observed.append(len(raw_emb) if isinstance(raw_emb, list) else 0)
+                except Exception:
+                    dims_observed.append(0)
+
+                # Validate and coerce embedding list
+                if not isinstance(raw_emb, list) or len(raw_emb) == 0:
+                    continue
+                try:
+                    # Ensure numeric float32 list
+                    clean_emb = [float(v) for v in raw_emb]
+                except Exception:
+                    continue
+
+                # Dimension consistency
+                if expected_dim is None:
+                    expected_dim = len(clean_emb)
+                if len(clean_emb) != expected_dim:
+                    # Collect a few examples to aid debugging
+                    try:
+                        mismatch_examples.append((str(item.get("staffId", "")), len(clean_emb)))
+                    except Exception:
+                        mismatch_examples.append(("", len(clean_emb)))
+                    self.logger.warning(f"Skipping embedding with mismatched dimension: got {len(clean_emb)} expected {expected_dim}")
+                    continue
+
                 staff_embedding = StaffEmbedding(
-                    embedding_id=item.get("embeddingId", ""),
-                    staff_id=item.get("staffId", ""),
-                    embedding=item.get("embedding", []),
-                    employee_id=str(item.get("employeeId", "")),
-                    staff_details=item.get("staffDetails", {}),
-                    is_active=item.get("isActive", True)
+                    embedding_id=(item.get("embeddingId", "") if isinstance(item, dict) else ""),
+                    staff_id=(item.get("staffId", "") if isinstance(item, dict) else ""),
+                    embedding=clean_emb,
+                    employee_id=str(item.get("employeeId", "")) if isinstance(item, dict) else "",
+                    staff_details=(item.get("staffDetails", {}) if isinstance(item, dict) else {}),
+                    is_active=(item.get("isActive", True) if isinstance(item, dict) else True)
                 )
-                
-                if staff_embedding.embedding:  # Only add if embedding exists
-                    self.staff_embeddings.append(staff_embedding)
-                    embeddings_list.append(staff_embedding.embedding)
+
+                self.staff_embeddings.append(staff_embedding)
+                embeddings_list.append(clean_emb)
             
             # Create numpy matrix for fast similarity computation (thread-safe)
             with self._embeddings_lock:
@@ -234,12 +269,27 @@ class EmbeddingManager:
                     
                     self.embedding_metadata = self.staff_embeddings.copy()
                     self.staff_embeddings_last_update = time.time()
-                    self.logger.info(f"Successfully loaded and cached {len(self.staff_embeddings)} staff embeddings")
-                    self.logger.debug(f"Embeddings matrix shape: {self.embeddings_matrix.shape}")
+                    self.logger.info(f"Successfully loaded and cached {len(self.staff_embeddings)} staff embeddings (dim={self.embeddings_matrix.shape[1]})")
+                    try:
+                        # Quick sanity metrics
+                        row0_sum = float(np.sum(self.embeddings_matrix[0])) if self.embeddings_matrix.shape[0] > 0 else 0.0
+                        self.logger.debug(f"Embeddings matrix shape: {self.embeddings_matrix.shape}, dtype={self.embeddings_matrix.dtype}, row0_sum={row0_sum:.4f}")
+                    except Exception:
+                        pass
                     return True
                 else:
-                    self.logger.warning("No active staff embeddings found in API response")
-                    return False
+                    # Build diagnostics and raise to stop pipeline early with actionable info
+                    dims_summary: Dict[int, int] = {}
+                    for d in dims_observed:
+                        dims_summary[d] = dims_summary.get(d, 0) + 1
+                    self.logger.error(
+                        "No valid staff embeddings loaded. Observed dimension distribution: "
+                        f"{dims_summary}. Expected_dim={expected_dim}. Mismatch examples (staffId, dim): "
+                        f"{mismatch_examples[:5]}"
+                    )
+                    raise RuntimeError(
+                        f"Failed to load staff embeddings due to dimension mismatch. Observed dims: {dims_summary}"
+                    )
                 
         except Exception as e:
             self.logger.error(f"Error loading staff embeddings: {e}", exc_info=True)
@@ -292,6 +342,10 @@ class EmbeddingManager:
             
         try:
             query_array = np.array(query_embedding, dtype=np.float32).reshape(1, -1)
+            # Dimension check
+            if embeddings_matrix.shape[1] != query_array.shape[1]:
+                self.logger.warning(f"Query embedding dim mismatch: query={query_array.shape[1]} staff={embeddings_matrix.shape[1]}")
+                return None
             
             # Normalize query embedding
             query_norm = np.linalg.norm(query_array)
@@ -327,6 +381,8 @@ class EmbeddingManager:
             return 0.0
         try:
             query_array = np.array(query_embedding, dtype=np.float32).reshape(1, -1)
+            if embeddings_matrix.shape[1] != query_array.shape[1]:
+                return 0.0
             qn = np.linalg.norm(query_array)
             if qn == 0:
                 return 0.0