Add option is_included for data_transform

This parameter can be used if the augmentations are already fully
specified within the dataset class.

tsimcne/tsimcne.py

@@ -180,11 +180,17 @@ class TSimCNE:
         then this needs to be appropriately reflected in the
         projection head as well.
 
-    :param None data_transform: The data augmentations to create the
-        differing views of the input.  By default it will use the same
-        augmentations as written in Böhm et al. (2023); random
-        cropping, greyscaling, color jitter, horizontal flips.  This
-        parameter should be changed with care.
+    :param None | ``"is_included"`` | torchvision.transforms |
+        ffcv.transforms data_transform: The data augmentations to
+        create the differing views of the input.  By default it will
+        use the same augmentations as written in Böhm et al. (2023);
+        random cropping, greyscaling, color jitter, horizontal flips.
+        This parameter should be changed with care.
+
+        If ``data_transform="is_included"``, then it is assumed that
+        all of the data augmentations are already included within the
+        dataset object ``X`` (which is passed to `tsimcne.fit`) and
+        will return the augmented samples along with a (dummy) label.
 
     :param [1000, 50, 450] total_epochs: A list of the number of
         epochs per training stage.  The ratio between the stages
@@ -445,7 +451,7 @@ def fit_transform(
             return_backbone_feat=return_backbone_feat,
         )
 
-    def fit(self, X: torch.utils.data.Dataset | str):
+    def fit(self, X: torch.utils.data.Dataset | str | Path):
         """Learn the mapping from the dataset ``X`` to 2D.
 
         :param X: The image dataset to be used for training.  Will be
@@ -462,7 +468,7 @@ def fit(self, X: torch.utils.data.Dataset | str):
                 self.use_ffcv = False
         self.check_ffcv(self.use_ffcv)
 
-        train_dl = self.make_dataloader(X, True, None)
+        train_dl = self.make_dataloader(X, True, self.data_transform)
 
         self.data_transform_none = get_transforms_unnormalized(
             size=self.image_size, setting="none", use_ffcv=self.use_ffcv
@@ -613,11 +619,17 @@ def make_dataloader(self, X, train_or_test, data_transform):
                 data_transform = get_transforms_unnormalized(
                     size=size, setting="none", use_ffcv=self.use_ffcv
                 )
+        else:
+            data_transform = self.data_transform
 
         if not self.use_ffcv:
-            # dataset that returns two augmented views of a given
-            # datapoint (and label)
-            dataset_contrastive = TransformedPairDataset(X, data_transform)
+            if data_transform != "is_included":
+                # dataset that returns two augmented views of a given
+                # datapoint (and label)
+                dataset_contrastive = TransformedPairDataset(X, data_transform)
+            else:
+                dataset_contrastive = X
+
             # wrap dataset into dataloader
             loader = torch.utils.data.DataLoader(
                 dataset_contrastive,