[IBR-2069] Modify hidden activation function (#317)

Author: isk03276

configs/lunarlander_continuous_v2/ppo.yaml

@@ -27,6 +27,7 @@ learner_cfg:
       type: "GaussianDist"
       configs: 
         hidden_sizes: [256, 256]
+        hidden_activation: "tanh"
         output_activation: "identity"
         fixed_logstd: True
     critic:

rl_algorithms/common/helper_functions.py

@@ -28,6 +28,11 @@ def identity(x: torch.Tensor) -> torch.Tensor:
     return x
 
 
+def relu(x: torch.Tensor) -> torch.Tensor:
+    """Return torch.relu(x)"""
+    return torch.relu(x)
+
+
 def soft_update(local: nn.Module, target: nn.Module, tau: float):
     """Soft-update: target = tau*local + (1-tau)*target."""
     for t_param, l_param in zip(target.parameters(), local.parameters()):

rl_algorithms/common/networks/heads.py

@@ -61,7 +61,11 @@ def __init__(
         self.hidden_sizes = configs.hidden_sizes
         self.input_size = configs.input_size
         self.output_size = configs.output_size
-        self.hidden_activation = hidden_activation
+        self.hidden_activation = (
+            getattr(helper_functions, configs.hidden_activation)
+            if "hidden_activation" in configs.keys()
+            else hidden_activation
+        )
         self.output_activation = getattr(helper_functions, configs.output_activation)
         self.linear_layer = linear_layer
         self.use_output_layer = use_output_layer