fix: 🚧 Subclass torch dataset Subset, hunting validation blocks bug

src/cellmap_data/multidataset.py

@@ -185,7 +185,7 @@ def get_indices(self, chunk_size: dict[str, int]) -> Sequence[int]:
         index_offset = 0
         for dataset in self.datasets:
             indices.append(dataset.get_indices(chunk_size))
-            index_offset += len(dataset)
+            index_offset += len(dataset) - 1
         return indices
 
     def set_raw_value_transforms(self, transforms: Callable):

src/cellmap_data/subdataset.py

@@ -1,18 +1,15 @@
-from typing import Callable
-from torch.utils.data import Dataset
+from typing import Callable, Sequence
+from torch.utils.data import Subset
+from .dataset import CellMapDataset
 
 
-class CellMapSubset(Dataset):
-    def __init__(self, dataset, indices):
-        super().__init__()
-        self.dataset = dataset
-        self.indices = indices
+class CellMapSubset(Subset):
 
-    def __getitem__(self, idx):
-        return self.dataset[self.indices[idx]]
+    dataset: CellMapDataset
+    indices: Sequence[int]
 
-    def __len__(self):
-        return len(self.indices)
+    def __init__(self, dataset: CellMapDataset, indices: Sequence[int]) -> None:
+        super().__init__(dataset, indices)
 
     @property
     def classes(self):