feat: add domain incremental ocl datasets (#344)

Author: tachyonicClock

docs/conf.py

@@ -55,6 +55,8 @@
     ("py:class", r"sklearn\..*"),
     ("py:class", r"torch\..*"),
     ("py:class", r"tqdm\..*"),
+    ("py:class", r"torchvision\..*"),
+    ("py:class", r"Tensor"),
 ]
 
 # These warnings are usually false positives.

docs/images/DomainCIFAR100.jpg

@@ -168,10 +168,12 @@ def __init__(
         data: torch.Tensor,
         targets: torch.Tensor,
         transform: Optional[Callable[[torch.Tensor], torch.Tensor]] = None,
+        target_transform: Optional[Callable[[object], object]] = None,
     ):
         self.data = data
         self.targets = targets
         self.transform = transform
+        self.target_transform = target_transform
 
     def __len__(self):
         return len(self.data)
@@ -182,5 +184,7 @@ def __getitem__(self, idx: int) -> Tuple[torch.Tensor, torch.Tensor]:
 
         if self.transform:
             x = self.transform(x)
+        if self.target_transform:
+            y = self.target_transform(y)
 
         return x, y

docs/images/RotatedFashionMNIST.jpg

@@ -34,19 +34,19 @@
 
 >>> with np.printoptions(precision=2):
 ...     print(metrics.accuracy_matrix)
-[[0.88 0.17 0.   0.05 0.08]
- [0.85 0.85 0.   0.05 0.08]
- [0.75 0.8  0.6  0.08 0.03]
- [0.73 0.77 0.52 0.38 0.1 ]
- [0.75 0.75 0.5  0.38 0.57]]
+[[0.88 0.17 0.08 0.05 0.08]
+ [0.85 0.85 0.03 0.05 0.08]
+ [0.75 0.77 0.6  0.08 0.03]
+ [0.75 0.77 0.52 0.38 0.1 ]
+ [0.75 0.75 0.52 0.38 0.57]]
 
 Notice that the accuracies in the upper triangle are close to zero because the
 learner has not trained on those tasks yet. The diagonal contains the accuracy
 on each task after training on that task. The lower triangle contains the
 accuracy on each task after training on all tasks.
 
 >>> print(f"Forward Transfer: {metrics.forward_transfer:0.2f}")
-Forward Transfer: 0.06
+Forward Transfer: 0.07
 
 >>> print(f"Backward Transfer: {metrics.backward_transfer:0.2f}")
 Backward Transfer: -0.08

docs/images/RotatedMNIST.jpg

@@ -0,0 +1,77 @@
+"""Use built-in datasets for online continual learning.
+
+In OCL datastreams are irreversible sequences of examples following a
+non-stationary data distribution. Learners in OCL can only learn from a single
+pass through the datastream but are expected to perform well on any portion of
+the datastream.
+
+Portions of the datastream where the data distribution is relatively stationary
+are called *tasks*.
+
+A common way to construct an OCL dataset for experimentation is to group the
+classes of a classification dataset into tasks. Known as the *class-incremental*
+scenario, the learner is presented with a sequence of tasks where each task
+contains a new subset of the classes.
+
+For example :class:`SplitMNIST` splits the MNIST dataset into five tasks where
+each task contains two classes:
+
+>>> from capymoa.ocl.datasets import SplitMNIST
+>>> scenario = SplitMNIST(preload_test=False)
+>>> scenario.task_schedule
+[{1, 4}, {5, 7}, {9, 3}, {0, 8}, {2, 6}]
+
+
+To get the usual CapyMOA stream object for training:
+
+>>> instance = scenario.stream.next_instance()
+>>> instance
+LabeledInstance(
+    Schema(SplitMNIST10/5),
+    x=[0. 0. 0. ... 0. 0. 0.],
+    y_index=4,
+    y_label='4'
+)
+
+CapyMOA streams flatten the data into a feature vector:
+
+>>> instance.x.shape
+(784,)
+
+You can access the PyTorch datasets for each task:
+
+>>> x, y = scenario.test_tasks[0][0]
+>>> x.shape
+torch.Size([1, 28, 28])
+>>> y
+1
+"""
+
+from ._base import _BuiltInCIScenario
+from ._tiny import RotatedTinyMNIST, TinySplitMNIST
+from ._vision import (
+    DomainCIFAR100,
+    RotatedFashionMNIST,
+    RotatedMNIST,
+    SplitCIFAR10,
+    SplitCIFAR100,
+    SplitFashionMNIST,
+    SplitMNIST,
+)
+from ._vit import DomainCIFAR100ViT, SplitCIFAR10ViT, SplitCIFAR100ViT
+
+__all__ = [
+    "_BuiltInCIScenario",
+    "SplitMNIST",
+    "RotatedMNIST",
+    "TinySplitMNIST",
+    "RotatedTinyMNIST",
+    "SplitCIFAR100ViT",
+    "SplitCIFAR10ViT",
+    "DomainCIFAR100ViT",
+    "SplitFashionMNIST",
+    "RotatedFashionMNIST",
+    "SplitCIFAR10",
+    "SplitCIFAR100",
+    "DomainCIFAR100",
+]