add Stan activation (PaddlePaddle#398)

ppsci/arch/activation.py

@@ -19,6 +19,27 @@
 from paddle import nn
 
 
+class Stan(nn.Layer):
+    """Self-scalable Tanh.
+    paper: https://arxiv.org/abs/2204.12589v1
+
+    Args:
+        out_features (int, optional): Output features. Defaults to 1.
+    """
+
+    def __init__(self, out_features: int = 1):
+        super().__init__()
+        self.beta = self.create_parameter(
+            shape=(out_features,),
+            default_initializer=nn.initializer.Constant(1),
+        )
+
+    def forward(self, x):
+        # TODO: manually broadcast beta to x.shape for preventing backward error yet.
+        return F.tanh(x) * (1 + paddle.broadcast_to(self.beta, x.shape) * x)
+        # return F.tanh(x) * (1 + self.beta * x)
+
+
 class Swish(nn.Layer):
     def __init__(self, beta: float = 1.0):
         super().__init__()
@@ -67,6 +88,7 @@ def forward(self, x):
     "swish": Swish(),
     "tanh": nn.Tanh(),
     "identity": nn.Identity(),
+    "stan": Stan,
 }
 
 

ppsci/arch/mlp.py

@@ -82,6 +82,7 @@ def __init__(
         self.input_keys = input_keys
         self.output_keys = output_keys
         self.linears = []
+        self.acts = []
         if isinstance(hidden_size, (tuple, list)):
             if num_layers is not None:
                 raise ValueError(
@@ -107,16 +108,18 @@ def __init__(
                 if weight_norm
                 else nn.Linear(cur_size, _size)
             )
+            # initialize activation function
+            self.acts.append(
+                act_mod.get_activation(activation)
+                if activation != "stan"
+                else act_mod.get_activation(activation)(_size)
+            )
             cur_size = _size
-        self.linears = nn.LayerList(self.linears)
 
+        self.linears = nn.LayerList(self.linears)
+        self.acts = nn.LayerList(self.acts)
         self.last_fc = nn.Linear(cur_size, len(self.output_keys))
 
-        # initialize activation function
-        self.act = nn.LayerList(
-            [act_mod.get_activation(activation) for _ in range(len(hidden_size))]
-        )
-
         self.skip_connection = skip_connection
 
     def forward_tensor(self, x):
@@ -130,7 +133,7 @@ def forward_tensor(self, x):
                     y = y + skip
                 else:
                     skip = y
-            y = self.act[i](y)
+            y = self.acts[i](y)
 
         y = self.last_fc(y)