Merge pull request #122 from ENSTA-U2IS-AI/dev

✨ Add UCI classification datasets, improve binary classification, add LS support for BCEWithLogitsLoss, remove dependencies

auto_tutorials_source/tutorial_corruption.py

@@ -125,6 +125,7 @@ def show_images(transforms):
 
 # %%
 # 4. Other Corruptions
+# ~~~~~~~~~~~~~~~~~~~~
 
 from torch_uncertainty.transforms.corruption import (
     Brightness, Contrast, Elastic, JPEGCompression, Pixelate)

docs/source/api.rst

@@ -326,6 +326,7 @@ Losses
     ConfidencePenaltyLoss
     KLDiv
     ELBOLoss
+    BCEWithLogitsLSLoss
 
 Post-Processing Methods
 -----------------------
@@ -379,14 +380,28 @@ Classification
     TinyImageNetDataModule
     ImageNetDataModule
 
+UCI Tabular Classification
+^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+.. autosummary::
+    :toctree: generated/
+    :nosignatures:
+    :template: class.rst
+
+    BankMarketingDataModule
+    DOTA2GamesDataModule
+    HTRU2DataModule
+    OnlineShoppersDataModule
+    SpamBaseDataModule
+
 Regression
 ^^^^^^^^^^
 .. autosummary::
     :toctree: generated/
     :nosignatures:
     :template: class.rst
 
-    UCIDataModule
+    UCIRegressionDataModule
 
 .. currentmodule:: torch_uncertainty.datamodules.segmentation
 
@@ -432,6 +447,25 @@ Classification
     TinyImageNetC
     OpenImageO
 
+
+UCI Tabular Classification
+^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+.. currentmodule:: torch_uncertainty.datasets.classification.uci
+
+
+.. autosummary::
+    :toctree: generated/
+    :nosignatures:
+    :template: class.rst
+
+    BankMarketing
+    DOTA2Games
+    HTRU2
+    OnlineShoppers
+    SpamBase
+
+
 Regression
 ^^^^^^^^^^
 

docs/source/conf.py

@@ -15,7 +15,7 @@
     f"{datetime.now().year!s}, Adrien Lafage and Olivier Laurent"
 )
 author = "Adrien Lafage and Olivier Laurent"
-release = "0.3.0"
+release = "0.3.1"
 
 # -- General configuration ---------------------------------------------------
 # https://www.sphinx-doc.org/en/master/usage/configuration.html#general-configuration

docs/source/quickstart.rst

@@ -89,7 +89,7 @@ CIFAR10 datamodule.
   from lightning.pytorch import TUTrainer
 
   dm = CIFAR10DataModule(root="data", batch_size=32)
-  trainer = TUTTrainer(gpus=1, max_epochs=100)
+  trainer = TUTrainer(gpus=1, max_epochs=100)
   trainer.fit(routine, dm)
   trainer.test(routine, dm)
 

docs/source/references.rst

@@ -266,6 +266,17 @@ For the conflictual loss, consider citing:
 * Authors: *Mohammed Fellaji, Frédéric Pennerath, Brieuc Conan-Guez, and Miguel Couceiro*
 * Paper: `ArXiv 2024 <https://arxiv.org/pdf/2407.12211>`__.
 
+Binary Cross-Entropy with Logits Loss with Label Smoothing
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+For the binary cross-entropy with logits loss with label smoothing, consider citing:
+
+**Rethinking the Inception Architecture for Computer Vision**
+
+* Authors: *Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, and Zbigniew Wojna*
+* Paper: `CVPR 2016 <https://arxiv.org/pdf/1512.00567.pdf>`__.
+
+
 Metrics
 -------
 

experiments/regression/uci_datasets/deep_ensemble.py

@@ -2,10 +2,10 @@
 
 from torch_uncertainty import cli_main, init_args
 from torch_uncertainty.baselines import DeepEnsemblesBaseline
-from torch_uncertainty.datamodules import UCIDataModule
+from torch_uncertainty.datamodules import UCIRegressionDataModule
 
 if __name__ == "__main__":
-    args = init_args(DeepEnsemblesBaseline, UCIDataModule)
+    args = init_args(DeepEnsemblesBaseline, UCIRegressionDataModule)
     if args.root == "./data/":
         root = Path(__file__).parent.absolute().parents[2]
     else:
@@ -15,7 +15,7 @@
 
     # datamodule
     args.root = str(root / "data")
-    dm = UCIDataModule(dataset_name="kin8nm", **vars(args))
+    dm = UCIRegressionDataModule(dataset_name="kin8nm", **vars(args))
 
     # model
     args.task = "regression"

experiments/regression/uci_datasets/mlp.py

@@ -3,7 +3,7 @@
 
 from torch_uncertainty import TULightningCLI
 from torch_uncertainty.baselines.regression import MLPBaseline
-from torch_uncertainty.datamodules import UCIDataModule
+from torch_uncertainty.datamodules import UCIRegressionDataModule
 
 
 class MLPCLI(TULightningCLI):
@@ -12,7 +12,7 @@ def add_arguments_to_parser(self, parser: LightningArgumentParser) -> None:
 
 
 def cli_main() -> MLPCLI:
-    return MLPCLI(MLPBaseline, UCIDataModule)
+    return MLPCLI(MLPBaseline, UCIRegressionDataModule)
 
 
 if __name__ == "__main__":

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "flit_core.buildapi"
 
 [project]
 name = "torch_uncertainty"
-version = "0.3.0"
+version = "0.3.1"
 authors = [
     { name = "ENSTA U2IS", email = "[email protected]" },
     { name = "Adrien Lafage", email = "[email protected]" },
@@ -28,20 +28,14 @@ classifiers = [
     "Programming Language :: Python :: 3.10",
     "Programming Language :: Python :: 3.11",
     "Programming Language :: Python :: 3.12",
-    "Programming Language :: Python :: 3.13",
     "Programming Language :: Python :: 3 :: Only",
 ]
 dependencies = [
     "timm",
     "lightning[pytorch-extra]>=2.0",
     "torchvision>=0.16",
-    "tensorboard",
     "einops",
-    "torchinfo",
-    "huggingface-hub",
-    "scikit-learn",
     "matplotlib",
-    "numpy",
     "rich>=10.2.2",
     "seaborn",
 ]
@@ -55,8 +49,10 @@ image = [
 ]
 tabular = ["pandas"]
 dev = [
+    "scikit-learn",
+    "huggingface-hub",
     "torch_uncertainty[image]",
-    "ruff==0.6.9",
+    "ruff==0.7.4",
     "pytest-cov",
     "pre-commit",
     "pre-commit-hooks",
@@ -74,6 +70,7 @@ all = [
     "laplace-torch",
     "glest==0.0.1a1",
     "scipy",
+    "tensorboard",
 ]
 
 [project.urls]
@@ -89,7 +86,7 @@ line-length = 80
 target-version = "py310"
 lint.extend-select = [
     "A",
-    "ARG", 
+    "ARG",
     "B",
     "C4",
     "D",
@@ -170,5 +167,5 @@ include = ["*/torch-uncertainty/*"]
 omit = ["*/tests/*", "*/datasets/*"]
 
 [tool.coverage.report]
-exclude_lines = ["coverage: ignore", "raise NotImplementedError"]
+exclude_lines = ["coverage: ignore", "raise NotImplementedError", "raise ImportError"]
 ignore_errors = true

tests/baselines/test_batched.py

@@ -1,6 +1,5 @@
 import torch
 from torch import nn
-from torchinfo import summary
 
 from torch_uncertainty.baselines.classification import (
     ResNetBaseline,
@@ -23,7 +22,6 @@ def test_batched_18(self):
             groups=1,
         )
 
-        summary(net)
         _ = net(torch.rand(1, 3, 32, 32))
 
     def test_batched_50(self):
@@ -38,7 +36,6 @@ def test_batched_50(self):
             groups=1,
         )
 
-        summary(net)
         _ = net(torch.rand(1, 3, 40, 40))
 
 
@@ -56,5 +53,4 @@ def test_batched(self):
             groups=1,
         )
 
-        summary(net)
         _ = net(torch.rand(1, 3, 32, 32))

tests/baselines/test_masked.py

@@ -1,7 +1,6 @@
 import pytest
 import torch
 from torch import nn
-from torchinfo import summary
 
 from torch_uncertainty.baselines.classification import (
     ResNetBaseline,
@@ -25,7 +24,6 @@ def test_masked_18(self):
             groups=1,
         )
 
-        summary(net)
         _ = net(torch.rand(1, 3, 32, 32))
 
     def test_masked_50(self):
@@ -41,7 +39,6 @@ def test_masked_50(self):
             groups=1,
         )
 
-        summary(net)
         _ = net(torch.rand(1, 3, 40, 40))
 
     def test_masked_errors(self):
@@ -87,5 +84,4 @@ def test_masked(self):
             groups=1,
         )
 
-        summary(net)
         _ = net(torch.rand(1, 3, 32, 32))

tests/baselines/test_mc_dropout.py

@@ -1,6 +1,5 @@
 import torch
 from torch import nn
-from torchinfo import summary
 
 from torch_uncertainty.baselines.classification import (
     ResNetBaseline,
@@ -24,7 +23,6 @@ def test_standard(self):
             style="cifar",
             groups=1,
         )
-        summary(net)
         net(torch.rand(1, 3, 32, 32))
 
 
@@ -42,7 +40,6 @@ def test_standard(self):
             style="cifar",
             groups=1,
         )
-        summary(net)
         net(torch.rand(1, 3, 32, 32))
 
 
@@ -61,7 +58,6 @@ def test_standard(self):
             groups=1,
             last_layer_dropout=True,
         )
-        summary(net)
         net(torch.rand(1, 3, 32, 32))
 
         net = VGGBaseline(

tests/baselines/test_mimo.py

@@ -1,6 +1,5 @@
 import torch
 from torch import nn
-from torchinfo import summary
 
 from torch_uncertainty.baselines.classification import (
     ResNetBaseline,
@@ -25,7 +24,6 @@ def test_mimo_50(self):
             groups=1,
         ).eval()
 
-        summary(net)
         _ = net(torch.rand(1, 3, 32, 32))
 
     def test_mimo_18(self):
@@ -42,7 +40,6 @@ def test_mimo_18(self):
             groups=2,
         ).eval()
 
-        summary(net)
         _ = net(torch.rand(1, 3, 40, 40))
 
 
@@ -62,5 +59,4 @@ def test_mimo(self):
             groups=1,
         ).eval()
 
-        summary(net)
         _ = net(torch.rand(1, 3, 32, 32))

tests/baselines/test_packed.py

@@ -1,7 +1,6 @@
 import pytest
 import torch
 from torch import nn
-from torchinfo import summary
 
 from torch_uncertainty.baselines.classification import (
     ResNetBaseline,
@@ -28,8 +27,6 @@ def test_packed_50(self):
             groups=1,
         )
 
-        summary(net)
-
         _ = net(torch.rand(1, 3, 32, 32))
 
     def test_packed_18(self):
@@ -46,7 +43,6 @@ def test_packed_18(self):
             groups=2,
         )
 
-        summary(net)
         _ = net(torch.rand(1, 3, 40, 40))
 
     def test_packed_exception(self):
@@ -95,7 +91,6 @@ def test_packed(self):
             groups=1,
         )
 
-        summary(net)
         _ = net(torch.rand(1, 3, 32, 32))
 
 
@@ -114,8 +109,6 @@ def test_packed(self):
             gamma=1,
             groups=1,
         )
-
-        summary(net)
         _ = net(torch.rand(2, 3, 32, 32))
 
 
@@ -133,5 +126,4 @@ def test_packed(self):
             alpha=2,
             gamma=1,
         )
-        summary(net)
         _ = net(torch.rand(1, 3))