improvements from JClinton to speed up loading offline data

lerobot/common/datasets/lerobot_dataset.py

@@ -611,11 +611,24 @@ class LeRobotDataset(torch.utils.data.Dataset):
         return query_timestamps
 
     def _query_hf_dataset(self, query_indices: dict[str, list[int]]) -> dict:
-        return {
-            key: torch.stack(self.hf_dataset.select(q_idx)[key])
-            for key, q_idx in query_indices.items()
-            if key not in self.meta.video_keys
-        }
+        # Step 1: Combine all unique indices
+        all_indices = sorted({idx for indices in query_indices.values() for idx in indices})
+
+        # Step 2: Select all required data at once
+        selected_dataset = self.hf_dataset.select(all_indices).to_dict()
+        selected_dataset = {key: torch.tensor(values) for key, values in selected_dataset.items()}
+
+        # Step 3: Map original indices to their positions in the selected dataset
+        index_map = {original_idx: i for i, original_idx in enumerate(all_indices)}
+
+        # Step 4: Build the result for each key
+        results = {}
+        for key, q_indices in query_indices.items():
+            if key not in self.meta.video_keys:
+                mapped_indices = [index_map[idx] for idx in q_indices]
+                results[key] = torch.stack([selected_dataset[key][i] for i in mapped_indices])
+
+        return results
 
     def _query_videos(self, query_timestamps: dict[str, list[float]], ep_idx: int) -> dict:
         """Note: When using data workers (e.g. DataLoader with num_workers>0), do not call this function