Release 0.13.0 (#961)

Preparation for release of version 0.13.0

Release text:

The new skorch release is here and it has some changes that will be exiting for
some users.

- First of all, you may have heard of the [PyTorch 2.0
  release](https://pytorch.org/get-started/pytorch-2.0/), which includes the
  option to compile the PyTorch module for better runtime performance. This
  skorch release allows you to pass `compile=True` when initializing the net to
  enable compilation.
- Support for training on multiple GPUs with the help of the
  [`accelerate`](https://huggingface.co/docs/accelerate/index) package has been
  improved by fixing some bugs and providing a dedicated [history
  class](https://skorch.readthedocs.io/en/latest/user/history.html#distributed-history).
  Our documentation contains more information on [what to consider when training
  on multiple
  GPUs](https://skorch.readthedocs.io/en/latest/user/huggingface.html#caution-when-using-a-multi-gpu-setup).
- If you have ever been frustrated with your neural net not training properly,
  you know how hard it can be to discover the underlying issue. Using the new
  [`SkorchDoctor`](https://skorch.readthedocs.io/en/latest/helper.html#skorch.helper.SkorchDoctor)
  class will simplify the diagnosis of underlying issues. Take a look at the
  accompanying
  [notebook](https://nbviewer.org/github/skorch-dev/skorch/blob/master/notebooks/Skorch_Doctor.ipynb)

Apart from that, a few bugs have been fixed and the included notebooks have been
updated to properly install requirements on Google Colab.

We are grateful for external contributors, many thanks to:

- Kshiteej K (kshitij12345)
- Muhammad Abdullah (abdulasiraj)
- Royi (RoyiAvital)
- Sawradip Saha (sawradip)
- y10ab1 (y10ab1)

Find below the list of all changes since v0.12.1 below:

### Added
- Add support for compiled PyTorch modules using the `torch.compile` function,
  introduced in [PyTorch 2.0
  release](https://pytorch.org/get-started/pytorch-2.0/), which can greatly
  improve performance on new GPU architectures; to use it, initialize your net
  with the `compile=True` argument, further compilation arguments can be
  specified using the dunder notation, e.g. `compile__dynamic=True`
- Add a class
  [`DistributedHistory`](https://skorch.readthedocs.io/en/latest/history.html#skorch.history.DistributedHistory)
  which should be used when training in a multi GPU setting (#955)
- `SkorchDoctor`: A helper class that assists in understanding and debugging the
  neural net training, see [this
  notebook](https://nbviewer.org/github/skorch-dev/skorch/blob/master/notebooks/Skorch_Doctor.ipynb)
  (#912)
- When using `AccelerateMixin`, it is now possible to prevent unwrapping of the
  modules by setting `unwrap_after_train=True` (#963)

### Fixed
- Fixed install command to work with recent changes in Google Colab (#928)
- Fixed a couple of bugs related to using non-default modules and criteria
  (#927)
- Fixed a bug when using `AccelerateMixin` in a multi-GPU setup (#947)
- `_get_param_names` returns a list instead of a generator so that subsequent
  error messages return useful information instead of a generator `repr` string
  (#925)
- Fixed a bug that caused modules to not be sufficiently unwrapped at the end of
  training when using `AccelerateMixin`, which could prevent them from being
  pickleable (#963)

CHANGES.md

@@ -7,11 +7,19 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Added
+
+### Changed
+
+### Fixed
+
+## [0.13.0] - 2023-05-17
+
 ### Added
 - Add support for compiled PyTorch modules using the `torch.compile` function, introduced in [PyTorch 2.0 release](https://pytorch.org/get-started/pytorch-2.0/), which can greatly improve performance on new GPU architectures; to use it, initialize your net with the `compile=True` argument, further compilation arguments can be specified using the dunder notation, e.g. `compile__dynamic=True`
 - Add a class [`DistributedHistory`](https://skorch.readthedocs.io/en/latest/history.html#skorch.history.DistributedHistory) which should be used when training in a multi GPU setting (#955)
 - `SkorchDoctor`: A helper class that assists in understanding and debugging the neural net training, see [this notebook](https://nbviewer.org/github/skorch-dev/skorch/blob/master/notebooks/Skorch_Doctor.ipynb) (#912)
-- When using `AccelerateMixin`, it is now possible to prevent unwrapping of the modules by setting `unwrap_after_train=True`
+- When using `AccelerateMixin`, it is now possible to prevent unwrapping of the modules by setting `unwrap_after_train=True` (#963)
 
 ### Changed
 
@@ -22,7 +30,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - `_get_param_names` returns a list instead of a generator so that subsequent
   error messages return useful information instead of a generator `repr`
   string (#925)
-- Fixed a bug that caused modules to not be sufficiently unwrapped at the end of training when using `AccelerateMixin`, which could prevent them from being pickleable
+- Fixed a bug that caused modules to not be sufficiently unwrapped at the end of training when using `AccelerateMixin`, which could prevent them from being pickleable (#963)
 
 ## [0.12.1] - 2022-11-18
 
@@ -306,3 +314,4 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 [0.11.0]: https://github.com/skorch-dev/skorch/compare/v0.10.0...v0.11.0
 [0.12.0]: https://github.com/skorch-dev/skorch/compare/v0.11.0...v0.12.0
 [0.12.1]: https://github.com/skorch-dev/skorch/compare/v0.12.0...v0.12.1
+[0.13.0]: https://github.com/skorch-dev/skorch/compare/v0.12.1...v0.13.0

VERSION

@@ -1 +1 @@
-0.12.1dev
+0.13.0

skorch/dataset.py

@@ -333,13 +333,3 @@ def __call__(self, dataset, y=None, groups=None):
     def __repr__(self):
         # pylint: disable=useless-super-delegation
         return super(ValidSplit, self).__repr__()
-
-# TODO remove in skorch 0.13
-class CVSplit(ValidSplit):
-    def __init__(self, *args, **kwargs):
-        warnings.warn(
-            f"{self.__class__.__name__} is deprecated, use the new name ValidSplit instead",
-            DeprecationWarning,
-            stacklevel=2,
-        )
-        super().__init__(*args, **kwargs)

skorch/net.py

@@ -1777,19 +1777,6 @@ def get_iterator(self, dataset, training=False):
           mini-batches.
 
         """
-        # TODO: remove in skorch v0.13, see #835
-        if isinstance(dataset, DataLoader):
-            msg = (
-                "get_iterator was called with a DataLoader instance but it should be "
-                "called with a Dataset instead. Probably, you implemented a custom "
-                "run_single_epoch method. Its first argument is now a DataLoader, "
-                "not a Dataset. For more information, look here: "
-                "https://skorch.readthedocs.io/en/latest/user/FAQ.html"
-                "#migration-from-0-11-to-0-12. This will raise an error in skorch v0.13"
-            )
-            warnings.warn(msg, DeprecationWarning)
-            return dataset
-
         if training:
             kwargs = self.get_params_for('iterator_train')
             iterator = self.iterator_train

skorch/tests/test_dataset.py

@@ -918,11 +918,3 @@ def test_random_state_not_used_raises(self, valid_split_cls):
 
     def test_random_state_and_float_does_not_raise(self, valid_split_cls):
         valid_split_cls(0.5, random_state=0)  # does not raise
-
-    def test_cvsplit_deprecation(self):
-        from skorch.dataset import CVSplit
-        with pytest.warns(
-            DeprecationWarning,
-            match="is deprecated, use the new name ValidSplit instead",
-        ):
-            CVSplit()

skorch/tests/test_net.py

@@ -3778,44 +3778,6 @@ def evaluation_step(self, batch, training=False):
         y_pred = net.predict(X)
         assert y_pred.shape == (100, 2)
 
-    # TODO: remove in skorch v0.13
-    def test_net_with_custom_run_single_epoch(self, net_cls, module_cls, data):
-        # See #835. We changed the API to initialize the DataLoader only once
-        # per epoch. This test is to make sure that code that overrides
-        # run_single_epoch still works for the time being.
-        from skorch.dataset import get_len
-
-        class MyNet(net_cls):
-            def run_single_epoch(self, dataset, training, prefix, step_fn, **fit_params):
-                # code as in skorch<=0.11
-                # first argument should now be an iterator, not a dataset
-                if dataset is None:
-                    return
-
-                # make sure that the "dataset" (really the DataLoader) can still
-                # access the Dataset if needed
-                assert hasattr(dataset, 'dataset')
-
-                batch_count = 0
-                for batch in self.get_iterator(dataset, training=training):
-                    self.notify("on_batch_begin", batch=batch, training=training)
-                    step = step_fn(batch, **fit_params)
-                    self.history.record_batch(prefix + "_loss", step["loss"].item())
-                    batch_size = (get_len(batch[0]) if isinstance(batch, (tuple, list))
-                                  else get_len(batch))
-                    self.history.record_batch(prefix + "_batch_size", batch_size)
-                    self.notify("on_batch_end", batch=batch, training=training, **step)
-                    batch_count += 1
-
-                self.history.record(prefix + "_batch_count", batch_count)
-
-        net = MyNet(module_cls, max_epochs=2)
-        X, y = data
-        with pytest.deprecated_call():
-            net.fit(X, y)
-        # does not raise
-        net.predict(X)
-
 
 class TestNetSparseInput:
     @pytest.fixture(scope='module')