implemented @Hartorn's feedback

examples/ex5_models_comparison.ipynb

@@ -153,15 +153,13 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 12,
+   "execution_count": 8,
    "metadata": {},
    "outputs": [
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "/Users/rak/Documents/loreal-poc/loreal_poc/tests/performance.py:47: RuntimeWarning: Mean of empty slice\n",
-      "  mes = np.nanmean(es, axis=1)\n",
       "/Users/rak/Documents/loreal-poc/loreal_poc/tests/performance.py:85: RuntimeWarning: Mean of empty slice\n",
       "  return np.nanmean(NMEs.get(prediction_result, marks))\n"
      ]
@@ -191,7 +189,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 13,
+   "execution_count": 9,
    "metadata": {},
    "outputs": [
     {
@@ -235,7 +233,7 @@
        "      <td>False</td>\n",
        "      <td>bottom half</td>\n",
        "      <td>OpenCV</td>\n",
-       "      <td>Cropped on bottom half</td>\n",
+       "      <td>300W cropped on bottom half</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>1</th>\n",
@@ -246,7 +244,7 @@
        "      <td>True</td>\n",
        "      <td>upper half</td>\n",
        "      <td>OpenCV</td>\n",
-       "      <td>Cropped on upper half</td>\n",
+       "      <td>300W cropped on upper half</td>\n",
        "    </tr>\n",
        "  </tbody>\n",
        "</table>\n",
@@ -257,12 +255,12 @@
        "0  TestDiff    NME_mean           NaN          1   False  bottom half   \n",
        "1  TestDiff    NME_mean      0.040216          1    True   upper half   \n",
        "\n",
-       "  model_name         dataloader_name  \n",
-       "0     OpenCV  Cropped on bottom half  \n",
-       "1     OpenCV   Cropped on upper half  "
+       "  model_name              dataloader_name  \n",
+       "0     OpenCV  300W cropped on bottom half  \n",
+       "1     OpenCV   300W cropped on upper half  "
       ]
      },
-     "execution_count": 13,
+     "execution_count": 9,
      "metadata": {},
      "output_type": "execute_result"
     }
@@ -290,7 +288,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.10.12"
+   "version": "3.11.6"
   }
  },
  "nbformat": 4,

loreal_poc/dataloaders/base.py

@@ -158,9 +158,12 @@ def _validate_image(cls, image: np.ndarray) -> None:
 
 
 class DataLoaderWrapper(DataIteratorBase):
-    def __init__(self, dataloader: DataIteratorBase, name: Optional[str] = None) -> None:
+    def __init__(self, dataloader: DataIteratorBase) -> None:
         self._wrapped_dataloader = dataloader
-        self.name = name
+
+    @property
+    def name(self):
+        return f"{self.__class__.__name__}({self._wrapped_dataloader.name})"
 
     def __len__(self) -> int:
         return len(self._wrapped_dataloader)

loreal_poc/dataloaders/wrappers.py

@@ -16,13 +16,16 @@ def __init__(
         crop_img: bool = True,
         crop_marks: bool = True,
     ) -> None:
-        name = f"Cropped on {part.name}"
-        super().__init__(dataloader, name=name)
+        super().__init__(dataloader)
         self._part = part
         self._margins = margins
         self.crop_img = crop_img
         self.crop_marks = crop_marks
 
+    @property
+    def name(self):
+        return f"{self._wrapped_dataloader.name} cropped on {self._part.name}"
+
     def get_image(self, idx: int) -> np.ndarray:
         image = super().get_image(idx)
         if not self.crop_img:
@@ -47,11 +50,15 @@ def __getitem__(self, idx: int) -> Tuple[np.ndarray, Optional[np.ndarray], Optio
 
 class CachedDataLoader(DataLoaderWrapper):
     def __init__(self, dataloader: DataIteratorBase, cache_size: int = 20) -> None:
-        super().__init__(dataloader, name="Cached")
+        super().__init__(dataloader)
         self._max_size: int = cache_size
         self._cache_idxs: List[int] = []
         self._cache: Dict[int, Tuple[np.ndarray, Optional[np.ndarray], Optional[Dict[Any, Any]]]] = {}
 
+    @property
+    def name(self):
+        return f"Cached {self._wrapped_dataloader.name}"
+
     def __getitem__(self, idx: int) -> Tuple[np.ndarray, Optional[np.ndarray], Optional[Dict[Any, Any]]]:
         # Add basic LRU cache to avoid reloading images and marks on small dataloaders
         if idx in self._cache_idxs: