Fix loss, comments

Author: BloodAxe

pytorch_toolbelt/modules/backbone/wider_resnet.py

@@ -18,7 +18,7 @@ def __init__(self,
                  groups=1,
                  norm_act=ABN,
                  dropout=None):
-        """Configurable identity-mapping residual block
+        """Identity-mapping residual block
         Parameters
         ----------
         in_channels : int
@@ -97,6 +97,7 @@ def __init__(self,
                  norm_act=ABN,
                  classes=0):
         """Wider ResNet with pre-activation (identity mapping) blocks
+
         Parameters
         ----------
         structure : list of int
@@ -168,8 +169,9 @@ def __init__(self,
                  norm_act=ABN,
                  classes=0,
                  dilation=False):
-        """Wider ResNet with pre-activation (identity mapping) blocks
+        """Wider ResNet with pre-activation (identity mapping) blocks.
         This variant uses down-sampling by max-pooling in the first two blocks and by strided convolution in the others.
+
         Parameters
         ----------
         structure : list of int

setup.cfg

@@ -1,4 +1,4 @@
 [flake8]
 max-line-length = 179
 exclude =.git,__pycache__,docs/source/conf.py,build,dist
-ignore = I101,I201,F401,F403,S001,D100,D101,D102,D103,D104,D105,D106,D107,D200,D205,D400,W504,E127
+ignore = I101,I201,F401,F403,S001,D100,D101,D102,D103,D104,D105,D106,D107,D200,D205,D400,D413,W504,E127

tests/test_losses.py

@@ -15,17 +15,13 @@ def test_sigmoid_focal_loss():
 
 
 def test_reduced_focal_loss():
-    input_bad = torch.Tensor([-1]).float()
-    input_good = torch.Tensor([10]).float()
-    target = torch.Tensor([1])
-
-    focal_loss_val = F.sigmoid_focal_loss(input_good, target)
-    reduced_loss_val = F.reduced_focal_loss(input_good, target)
-    assert reduced_loss_val < focal_loss_val
+    input_good = torch.Tensor([10, -10, 10]).float()
+    input_bad = torch.Tensor([-1, 2, 0]).float()
+    target = torch.Tensor([1, 0, 1])
 
-    focal_loss_val = F.sigmoid_focal_loss(input_bad, target)
-    reduced_loss_val = F.reduced_focal_loss(input_bad, target)
-    assert reduced_loss_val == focal_loss_val
+    loss_good = F.reduced_focal_loss(input_good, target)
+    loss_bad = F.reduced_focal_loss(input_bad, target)
+    assert loss_good < loss_bad
 
 
 def test_soft_jaccard_score():