update version with file + cleaning

.github/ISSUE_TEMPLATE/generic_bug_report.yaml

@@ -58,7 +58,7 @@ body:
     description: |
       examples:
         - **OS**: linux
-        - **Python version**: 3.7
+        - **Python version**: 3.11
         - **Packages used version**: PyTorch, Numpy, scikit-learn, etc..
     value: |
         - OS:

MANIFEST.in

@@ -0,0 +1,2 @@
+include deel/torchlip/VERSION
+include LICENSE

deel/torchlip/VERSION

@@ -0,0 +1 @@
+0.2

deel/torchlip/__init__.py

@@ -26,7 +26,10 @@
 # =====================================================================================
 # flake8: noqa
 
-__version__ = "0.1.1"
+from os import path
+
+with open(path.join(path.dirname(__file__), "VERSION")) as f:
+    __version__ = f.read().strip()
 
 from . import functional, init, normalizers, utils
 from .modules import *

deel/torchlip/functional.py

@@ -404,33 +404,6 @@ def neg_kr_loss(
     return -kr_loss(input, target, multi_gpu=multi_gpu)
 
 
-# def neg_kr_loss_multi_gpu(
-#     input: torch.Tensor,
-#     target: torch.Tensor,
-# ) -> torch.Tensor:
-#     """
-#     Loss to estimate the negative wasserstein-1 distance using Kantorovich-Rubinstein
-#     duality.
-
-#     `target` and `input` can be either of shape (batch_size, 1) or
-#     (batch_size, # classes).
-
-#     When using this loss function, the labels `target` must be pre-processed with the
-#     `process_labels_for_multi_gpu()` function.
-
-#     Args:
-#         input: Tensor of arbitrary shape.
-#         target: pre-processed Tensor of the same shape as input.
-
-#     Returns:
-#         The negative Wasserstein-1 loss between ``input`` and ``target``.
-
-#     See Also:
-#         :py:func:`kr_loss`
-#     """
-#     return -kr_loss_multi_gpu(input, target)
-
-
 def hinge_margin_loss(
     input: torch.Tensor,
     target: torch.Tensor,
@@ -497,69 +470,6 @@ def hkr_loss(
     ) * kr_loss_fct(input, target)
 
 
-# def hkr_loss_multi_gpu(
-#     input: torch.Tensor,
-#     target: torch.Tensor,
-#     alpha: float,
-#     min_margin: float = 1.0,
-# ) -> torch.Tensor:
-#     """
-#     Loss to estimate the wasserstein-1 distance with a hinge regularization using
-#     Kantorovich-Rubinstein duality.
-
-#     Args:
-#         input: Tensor of arbitrary shape.
-#         target: Tensor of the same shape as input.
-#         alpha: Regularization factor between the hinge and the KR loss.
-#         min_margin: Minimal margin for the hinge loss.
-#         true_values: tuple containing the two label for each predicted class.
-
-#     Returns:
-#         The regularized Wasserstein-1 loss.
-
-#     See Also:
-#         :py:func:`hinge_margin_loss`
-#         :py:func:`kr_loss`
-#     """
-#     assert alpha <= 1.0
-#     if alpha == 1.0:  # alpha for  hinge only
-#         return hinge_margin_loss(input, target, min_margin)
-#     if alpha == 0:
-#         return -kr_loss_multi_gpu(input, target)
-#     # true value: positive value should be the first to be coherent with the
-#     # hinge loss (positive y_pred)
-#     return alpha * hinge_margin_loss(input, target, min_margin) - (
-#         1 - alpha
-#     ) * kr_loss_multi_gpu(input, target)
-
-
-# def kr_multiclass_loss(
-#     input: torch.Tensor,
-#     target: torch.Tensor,
-# ) -> torch.Tensor:
-#     r"""
-#     Loss to estimate average of W1 distance using Kantorovich-Rubinstein
-#     duality over outputs. In this multiclass setup thr KR term is computed
-#     for each class and then averaged.
-
-#     Args:
-#         input: Tensor of arbitrary shape.
-#         target: Tensor of the same shape as input.
-#                 target has to be one hot encoded (labels being 1s and 0s ).
-
-#     Returns:
-#         The Wasserstein multiclass loss between ``input`` and ``target``.
-#     """
-#     return kr_loss(input, target)
-#     # true_target = torch.where(target > 0, 1.0, 0.0).to(input.dtype)
-#     # esp_true_true = torch.sum(input * true_target, 0) / torch.sum(true_target, 0)
-#     # esp_false_true = torch.sum(input * (1 - true_target), 0) / torch.sum(
-#     #     (1 - true_target), 0
-#     # )
-
-#     # return torch.mean(esp_true_true - esp_false_true)
-
-
 def hinge_multiclass_loss(
     input: torch.Tensor,
     target: torch.Tensor,
@@ -629,42 +539,6 @@ def hkr_multiclass_loss(
         ) * kr_loss_fct(input, target)
 
 
-# def hkr_multiclass_loss_multi_gpu(
-#     input: torch.Tensor,
-#     target: torch.Tensor,
-#     alpha: float = 0.0,
-#     min_margin: float = 1.0,
-# ) -> torch.Tensor:
-#     """
-#     Loss to estimate the wasserstein-1 distance with a hinge regularization using
-#     Kantorovich-Rubinstein duality.
-
-#     Args:
-#         input: Tensor of arbitrary shape.
-#         target: Tensor of the same shape as input.
-#         alpha: Regularization factor between the hinge and the KR loss.
-#         min_margin: Minimal margin for the hinge loss.
-#         true_values: tuple containing the two label for each predicted class.
-
-#     Returns:
-#         The regularized Wasserstein-1 loss.
-
-#     See Also:
-#         :py:func:`hinge_margin_loss`
-#         :py:func:`kr_loss`
-#     """
-
-#     assert alpha <= 1.0
-#     if alpha == 1.0:  # alpha  hinge only
-#         return hinge_multiclass_loss(input, target, min_margin)
-#     elif alpha == 0.0:  # alpha = 0 => KR only
-#         return -kr_loss_multi_gpu(input, target)
-#     else:
-#         return alpha * hinge_multiclass_loss(input, target, min_margin) - (
-#             1 - alpha
-#         ) * kr_loss_multi_gpu(input, target)
-
-
 def process_labels_for_multi_gpu(labels: torch.Tensor) -> torch.Tensor:
     """Process labels to be fed to any loss based on KR estimation with a multi-GPU/TPU
     strategy.

setup.py

@@ -34,6 +34,9 @@
 with open(path.join(this_directory, "README.md"), encoding="utf-8") as f:
     long_description = f.read()
 
+with open(path.join(this_directory, "deel/torchlip/VERSION")) as f:
+    version = f.read().strip()
+
 dev_requires = ["tox", "black", "flake8", "flake8-black", "numpy", "torch_testing"]
 
 docs_requires = [
@@ -49,7 +52,7 @@
 
 setuptools.setup(
     name="deel-torchlip",
-    version="0.1.1",
+    version=version,
     author=", ".join(
         [
             "Mathieu SERRURIER",
@@ -74,7 +77,6 @@
     packages=setuptools.find_namespace_packages(include=["deel.*"]),
     install_requires=[
         "numpy",
-        "inflection",
         "torch",
     ],
     license="MIT",

tests/test_models.py

@@ -305,8 +305,6 @@ def test_warning_unsupported_1Lip_layers():
             tActivation, {"activation": "gelu"}
         ),  # kl.Activation("relu"),
     ]
-    # if version.parse(tf.__version__) >= version.parse("2.4.0"):
-    #         unsupported_layers.append(kl.Activation("gelu"))
 
     for lay in unsupported_layers:
         if lay is not None: