fix: links in docstrings (#859)

catalyst/contrib/data/transforms.py

@@ -1,6 +1,7 @@
 """
-This subpackage was borrowed from
-torchvision(https://github.com/pytorch/vision).
+This subpackage was borrowed from `torchvision`__.
+
+__ https://github.com/pytorch/vision
 """
 
 import numpy as np

catalyst/contrib/nn/criterion/circle.py

@@ -25,9 +25,8 @@ def _convert_label_to_similarity(
 
 class CircleLoss(nn.Module):
     """
-    CircleLoss from
-    "Circle Loss: A Unified Perspective of Pair Similarity Optimization"
-    https://arxiv.org/abs/2002.10857
+    CircleLoss from `Circle Loss: A Unified Perspective
+    of Pair Similarity Optimization`_ paper.
 
     Adapter from:
     https://github.com/TinyZeaMays/CircleLoss
@@ -41,6 +40,9 @@ class CircleLoss(nn.Module):
         >>> labels = torch.randint(high=10, size=(256,))
         >>> criterion = CircleLoss(margin=0.25, gamma=256)
         >>> criterion(features, labels)
+
+    .. _`Circle Loss: A Unified Perspective of Pair Similarity Optimization`:
+        https://arxiv.org/abs/2002.10857
     """
 
     def __init__(self, margin: float, gamma: float) -> None:

catalyst/contrib/nn/criterion/triplet.py

@@ -7,8 +7,7 @@
 class TripletLoss(nn.Module):
     """Triplet loss with hard positive/negative mining.
 
-    Reference:
-        Code imported from https://github.com/NegatioN/OnlineMiningTripletLoss
+    Adapted from: https://github.com/NegatioN/OnlineMiningTripletLoss
     """
 
     def __init__(self, margin: float = 0.3):

catalyst/contrib/nn/modules/se.py

@@ -6,7 +6,7 @@
 class cSE(nn.Module):  # noqa: N801
     """
     The channel-wise SE (Squeeze and Excitation) block from the
-    [Squeeze-and-Excitation Networks](https://arxiv.org/abs/1709.01507) paper.
+    `Squeeze-and-Excitation Networks`__ paper.
 
     Adapted from
     https://www.kaggle.com/c/tgs-salt-identification-challenge/discussion/65939
@@ -17,6 +17,8 @@ class cSE(nn.Module):  # noqa: N801
 
     - Input: (batch, channels, height, width)
     - Output: (batch, channels, height, width) (same shape as input)
+
+    __ https://arxiv.org/abs/1709.01507
     """
 
     def __init__(self, in_channels: int, r: int = 16):
@@ -48,8 +50,8 @@ def forward(self, x: torch.Tensor):
 class sSE(nn.Module):  # noqa: N801
     """
     The sSE (Channel Squeeze and Spatial Excitation) block from the
-    [Concurrent Spatial and Channel ‘Squeeze & Excitation’
-    in Fully Convolutional Networks](https://arxiv.org/abs/1803.02579) paper.
+    `Concurrent Spatial and Channel ‘Squeeze & Excitation’
+    in Fully Convolutional Networks`__ paper.
 
     Adapted from
     https://www.kaggle.com/c/tgs-salt-identification-challenge/discussion/66178
@@ -58,6 +60,8 @@ class sSE(nn.Module):  # noqa: N801
 
     - Input: (batch, channels, height, width)
     - Output: (batch, channels, height, width) (same shape as input)
+
+    __ https://arxiv.org/abs/1803.02579
     """
 
     def __init__(self, in_channels: int):
@@ -83,8 +87,8 @@ def forward(self, x: torch.Tensor):
 class scSE(nn.Module):  # noqa: N801
     """
     The scSE (Concurrent Spatial and Channel Squeeze and Channel Excitation)
-    block from the [Concurrent Spatial and Channel ‘Squeeze & Excitation’
-    in Fully Convolutional Networks](https://arxiv.org/abs/1803.02579) paper.
+    block from the `Concurrent Spatial and Channel ‘Squeeze & Excitation’
+    in Fully Convolutional Networks`__ paper.
 
     Adapted from
     https://www.kaggle.com/c/tgs-salt-identification-challenge/discussion/66178
@@ -93,6 +97,8 @@ class scSE(nn.Module):  # noqa: N801
 
     - Input: (batch, channels, height, width)
     - Output: (batch, channels, height, width) (same shape as input)
+
+    __ https://arxiv.org/abs/1803.02579
     """
 
     def __init__(self, in_channels: int, r: int = 16):

catalyst/data/sampler.py

@@ -292,8 +292,7 @@ class DynamicLenBatchSampler(BatchSampler):
     A dynamic batch length data sampler.
     Should be used with `catalyst.utils.trim_tensors`.
 
-    Adapted from "Dynamic minibatch trimming to improve BERT training speed"
-    https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/discussion/94779
+    Adapted from `Dynamic minibatch trimming to improve BERT training speed`_.
 
     Args:
         sampler (torch.utils.data.Sampler): Base sampler.
@@ -321,6 +320,9 @@ class DynamicLenBatchSampler(BatchSampler):
         >>>     tensors = utils.trim_tensors(batch)
         >>>     b_input_ids, b_input_mask, b_segment_ids, b_labels = \
         >>>         tuple(t.to(device) for t in tensors)
+
+    .. _`Dynamic minibatch trimming to improve BERT training speed`:
+        https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/discussion/94779
     """
 
     def __iter__(self):

catalyst/utils/torch.py

@@ -330,14 +330,16 @@ def trim_tensors(tensors):
     Trim padding off of a batch of tensors to the smallest possible length.
     Should be used with `catalyst.data.DynamicLenBatchSampler`.
 
-    Adapted from "Dynamic minibatch trimming to improve BERT training speed"
-    https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/discussion/94779
+    Adapted from `Dynamic minibatch trimming to improve BERT training speed`_.
 
     Args:
         tensors ([torch.tensor]): list of tensors to trim.
 
     Returns:
         List[torch.tensor]: list of trimmed tensors.
+
+    .. _`Dynamic minibatch trimming to improve BERT training speed`:
+        https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/discussion/94779
     """
     max_len = torch.max(torch.sum((tensors[0] != 0), 1))
     if max_len > 2: