changing default network, making deepsets network choose output dimension appropriately, making bc config to be called by ray

config/networks.json5

@@ -1,32 +1,32 @@
 {
     ego_state: {
         input_dim: 5, // number of state vars
-        hidden_n: 1,
+        hidden_n: 0,
         hidden_dim: 5,
         output_dim: 5
     },
     deepsets: {
         input_dim: 5, // number of relative state vars for others
         phi: {
-            hidden_n: 1,
+            hidden_n: 2,
             hidden_dim: 20,
         },
         latent_dim: 20,
         rho: {
-            hidden_n: 1,
+            hidden_n: 2,
             hidden_dim: 10,
         },
         output_dim: 10
     },
     path_encoder: {
         input_dim: 40, // 2 * path length for (x,y) coordinates
-        hidden_n: 2,
+        hidden_n: 0,
         hidden_dim: 20,
         output_dim: 10,
     },
     head: {
         input_dim: 0, // computed in policy constructor
-        hidden_n: 1,
+        hidden_n: 3,
         hidden_dim: 50,
         output_dim: 1, // number of outputs e.g. number of actions, or just one
         final_activation: 'sigmoid',

src/bc/bc.py

@@ -8,6 +8,40 @@ from src.policies import DeepSetsPolicy
 from src.util.transform import MinMaxScaler
 from tqdm import tqdm
 
+def bc_config(ray_config):
+    config = {
+        'ego_state': {'input_dim': 5, 'hidden_n': 0, 'output_dim': 0},
+        'deepsets': {
+            'input_dim': 5,
+            'phi': {
+                'hidden_n': ray_config['deepsets_phi_hidden_n'],
+                'hidden_dim': ray_config['deepsets_phi_hidden_dim']
+                },
+            'latent_dim': ray_config['deepsets_latent_dim'],
+            'rho': {'hidden_n': 0, 'hidden_dim': 10},
+            'output_dim': 0
+        },
+        'path_encoder': {'input_dim': 40, 'hidden_n': 0, 'output_dim': 0},
+        'head': {
+            'input_dim': 0, # computed in constructor
+            'hidden_n': ray_config['head_hidden_n'],
+            'hidden_dim': ray_config['head_hidden_dim'],
+            'output_dim': 1, # number of outputs e.g. number of actions, or just one
+            'final_activation': ray_config['head_final_activation'],
+        },
+        'optim': {
+            'optimizer':'adam',
+            'lr':ray_config['lr'],
+            'weight_decay':ray_config['weight_decay']
+        },
+        'train_epochs':1000,
+        'train_batch_size': ray_config['batch_size'],
+        'loss': ray_config['loss'],
+
+    }
+    return config
+
+
 class BehaviorCloningPolicy():
     """
     Class for (continuous) behavior cloning policy

src/main.py

@@ -18,7 +18,7 @@ def basestr(**kwargs):
     Returns:
         basestr (str): prefix
     """
-    return 'base_'
+    return 'base'
 
 def main(method='bc', train=False, test=False, loc=0, config_path=None, **kwargs):
     """

src/nets/deepsets.py

@@ -18,9 +18,9 @@ class DeepSetsModule(nn.Module):
         super(DeepSetsModule, self).__init__()
         self.input_dim = input_dim
         self.latent_dim = latent_dim
-        self.output_dim = output_dim
         self.phi = Phi(self.input_dim, phi_hidden_n, phi_hidden_dim, self.latent_dim)
-        self.rho = Phi(self.latent_dim, rho_hidden_n, rho_hidden_dim, self.output_dim)
+        self.rho = Phi(self.latent_dim, rho_hidden_n, rho_hidden_dim, output_dim)
+        self.output_dim = self.rho.output_dim
         self.pooling = torch.sum
 
     @staticmethod