__call__ -> forward

labml_nn/resnet/__init__.py

@@ -83,7 +83,7 @@ class ShortcutProjection(Module):
         # Paper suggests adding batch normalization after each convolution operation
         self.bn = nn.BatchNorm2d(out_channels)
 
-    def __call__(self, x: torch.Tensor):
+    def forward(self, x: torch.Tensor):
         # Convolution and batch normalization
         return self.bn(self.conv(x))
 
@@ -140,7 +140,7 @@ class ResidualBlock(Module):
         # Second activation function (ReLU) (after adding the shortcut)
         self.act2 = nn.ReLU()
 
-    def __call__(self, x: torch.Tensor):
+    def forward(self, x: torch.Tensor):
         """
         * `x` is the input of shape `[batch_size, in_channels, height, width]`
         """
@@ -221,7 +221,7 @@ class BottleneckResidualBlock(Module):
         # Second activation function (ReLU) (after adding the shortcut)
         self.act3 = nn.ReLU()
 
-    def __call__(self, x: torch.Tensor):
+    def forward(self, x: torch.Tensor):
         """
         * `x` is the input of shape `[batch_size, in_channels, height, width]`
         """
@@ -310,7 +310,7 @@ class ResNetBase(Module):
         # Stack the blocks
         self.blocks = nn.Sequential(*blocks)
 
-    def __call__(self, x: torch.Tensor):
+    def forward(self, x: torch.Tensor):
         """
         * `x` has shape `[batch_size, img_channels, height, width]`
         """

labml_nn/rl/dqn/__init__.py

@@ -106,7 +106,7 @@ class QFuncLoss(Module):
         self.gamma = gamma
         self.huber_loss = nn.SmoothL1Loss(reduction='none')
 
-    def __call__(self, q: torch.Tensor, action: torch.Tensor, double_q: torch.Tensor,
+    def forward(self, q: torch.Tensor, action: torch.Tensor, double_q: torch.Tensor,
                 target_q: torch.Tensor, done: torch.Tensor, reward: torch.Tensor,
                 weights: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:
         """

labml_nn/rl/dqn/model.py

@@ -82,7 +82,7 @@ class Model(Module):
             nn.Linear(in_features=256, out_features=4),
         )
 
-    def __call__(self, obs: torch.Tensor):
+    def forward(self, obs: torch.Tensor):
         # Convolution
         h = self.conv(obs)
         # Reshape for linear layers

labml_nn/rl/ppo/__init__.py

@@ -136,7 +136,7 @@ class ClippedPPOLoss(Module):
     def __init__(self):
         super().__init__()
 
-    def __call__(self, log_pi: torch.Tensor, sampled_log_pi: torch.Tensor,
+    def forward(self, log_pi: torch.Tensor, sampled_log_pi: torch.Tensor,
                 advantage: torch.Tensor, clip: float) -> torch.Tensor:
         # ratio $r_t(\theta) = \frac{\pi_\theta (a_t|s_t)}{\pi_{\theta_{OLD}} (a_t|s_t)}$;
         # *this is different from rewards* $r_t$.
@@ -200,7 +200,8 @@ class ClippedValueFunctionLoss(Module):
      significantly from $V_{\theta_{OLD}}$.
 
     """
-    def __call__(self, value: torch.Tensor, sampled_value: torch.Tensor, sampled_return: torch.Tensor, clip: float):
+
+    def forward(self, value: torch.Tensor, sampled_value: torch.Tensor, sampled_return: torch.Tensor, clip: float):
         clipped_value = sampled_value + (value - sampled_value).clamp(min=-clip, max=clip)
         vf_loss = torch.max((value - sampled_return) ** 2, (clipped_value - sampled_return) ** 2)
         return 0.5 * vf_loss.mean()

labml_nn/rl/ppo/experiment.py

@@ -69,7 +69,7 @@ class Model(Module):
         #
         self.activation = nn.ReLU()
 
-    def __call__(self, obs: torch.Tensor):
+    def forward(self, obs: torch.Tensor):
         h = self.activation(self.conv1(obs))
         h = self.activation(self.conv2(h))
         h = self.activation(self.conv3(h))