Updated adam to reduce plasticity and made the agent more observant, amongst other things

.gitignore

@@ -164,3 +164,5 @@ cython_debug/
 
 # Random stuff
 __pycache__/
+chkpts/
+figures/

entities/player.py

@@ -86,8 +86,8 @@ class Player(pygame.sprite.Sprite):
             print("Attempting to load models ...")
             try:
                 self.agent.load_models(
-                    actr_chkpt=f"run{load}/A{self.player_id}",
+                    actr_chkpt=f"{chkpt_dir}/../run{load}/A{self.player_id}",
-                    crtc_chkpt=f"run{load}/C{self.player_id}"
+                    crtc_chkpt=f"{chkpt_dir}/../run{load}/C{self.player_id}"
                 )
                 print("Models loaded ...\n")
 
@@ -160,9 +160,9 @@ class Player(pygame.sprite.Sprite):
         if hasattr(self, 'state_features'):
             self.old_state_features = self.state_features
 
-            self.reward = self.stats.exp\
+            self.reward = self.stats.exp
-                + self.stats.health/self.stats.stats['health'] - 1\
+                # + self.stats.health/self.stats.stats['health'] - 1\
-                - nearest_dist/np.sqrt(np.sum(self.map_edge))
+                # - nearest_dist/np.sqrt(np.sum(self.map_edge))
 
         self.state_features = [
             self.animation.rect.center[0]/self.map_edge[0],
@@ -173,27 +173,27 @@ class Player(pygame.sprite.Sprite):
             self.stats.energy/self.stats.stats['energy'],
         ]
 
-        self.state_features.extend([
+        # self.state_features.extend([
-            nearest_dist/np.sqrt(np.sum(self.map_edge)),
+        #     nearest_dist/np.sqrt(np.sum(self.map_edge)),
-            nearest_en_dir[0],
+        #     nearest_en_dir[0],
-            nearest_en_dir[1],
+        #     nearest_en_dir[1],
-            nearest_enemy.stats.exp
+        #     nearest_enemy.stats.exp
-        ])
+        # ])
 
-        # for distance, direction, enemy in self.distance_direction_from_enemy:
+         for distance, direction, enemy in sorted_distances[:5]:
-        #     self.state_features.extend([
+             self.state_features.extend([
-        #
+        
-        #         distance/np.sqrt(np.sum(self.map_edge)),
+                 distance/np.sqrt(np.sum(self.map_edge)),
-        #
+        
-        #         direction[0],
+                 direction[0],
-        #
+        
-        #         direction[1],
+                 direction[1],
-        #
+        
-        #         enemy.stats.health /
+                 enemy.stats.health /
-        #         enemy.stats.monster_info['health'],
+                 enemy.stats.monster_info['health'],
-        #
+        
-        #         enemy.stats.exp,
+                 enemy.stats.exp,
-        #     ])
+             ])
 
         if hasattr(self, 'num_features'):
             while len(self.state_features) < self.num_features:

ml/ppo/brain.py

@@ -52,7 +52,7 @@ class PPOMemory:
 
 class ActorNetwork(nn.Module):
 
-    def __init__(self, input_dim, output_dim, alpha, fc1_dims=1024, fc2_dims=1024, chkpt_dir='tmp/ppo'):
+    def __init__(self, input_dim, output_dim, alpha, fc1_dims=512, fc2_dims=512, chkpt_dir='tmp'):
         super(ActorNetwork, self).__init__()
 
         self.chkpt_dir = chkpt_dir
@@ -68,7 +68,7 @@ class ActorNetwork(nn.Module):
             nn.Softmax(dim=-1)
         )
 
-        self.optimizer = optim.Adam(self.parameters(), lr=alpha, eps=1e-5)
+        self.optimizer = optim.Adam(self.parameters(), lr=alpha, betas=(0.9, 0.9), eps=1e-5)
 
         self.device = T.device('cuda:0' if T.cuda.is_available() else 'cpu')
 
@@ -80,18 +80,19 @@ class ActorNetwork(nn.Module):
 
         return dist
 
-    def save_checkpoint(self, filename='actor_torch_ppo'):
+    def save_checkpoint(self, filename):
-        T.save(self.state_dict(), os.path.join(self.chkpt_dir, filename))
+        T.save(self.state_dict(), os.path.join(filename))
 
-    def load_checkpoint(self, filename='actor_torch_ppo'):
+    def load_checkpoint(self, filename):
+        print(filename)
         self.load_state_dict(
-            T.load(os.path.join(self.chkpt_dir, filename),
+            T.load(os.path.join(filename),
                    map_location=self.device))
 
 
 class CriticNetwork(nn.Module):
 
-    def __init__(self, input_dims, alpha, fc1_dims=4096, fc2_dims=4096, chkpt_dir='tmp/ppo'):
+    def __init__(self, input_dims, alpha, fc1_dims=2048, fc2_dims=2048, chkpt_dir='tmp'):
         super(CriticNetwork, self).__init__()
 
         self.chkpt_dir = chkpt_dir
@@ -105,16 +106,16 @@ class CriticNetwork(nn.Module):
             nn.LeakyReLU(),
             nn.Linear(fc1_dims, fc2_dims),
             nn.LeakyReLU(),
-            nn.Linear(fc1_dims, fc2_dims),
+            # nn.Linear(fc1_dims, fc2_dims),
-            nn.LeakyReLU(),
+            # nn.LeakyReLU(),
-            nn.Linear(fc1_dims, fc2_dims),
+            # nn.Linear(fc1_dims, fc2_dims),
-            nn.LeakyReLU(),
+            # nn.LeakyReLU(),
             nn.Linear(fc1_dims, fc2_dims),
             nn.LeakyReLU(),
             nn.Linear(fc2_dims, 1)
         )
 
-        self.optimizer = optim.Adam(self.parameters(), lr=alpha, eps=1e-5)
+        self.optimizer = optim.Adam(self.parameters(), lr=alpha, betas=(0.9, 0.9), eps=1e-5)
 
         self.device = T.device('cuda:0' if T.cuda.is_available() else 'cpu')
 
@@ -124,10 +125,11 @@ class CriticNetwork(nn.Module):
         value = self.critic(state)
         return value
 
-    def save_checkpoint(self, filename='critic_torch_ppo'):
+    def save_checkpoint(self, filename):
-        T.save(self.state_dict(), os.path.join(self.chkpt_dir, filename))
+        T.save(self.state_dict(), os.path.join(filename))
 
-    def load_checkpoint(self, filename='critic_torch_ppo'):
+    def load_checkpoint(self, filename):
+        print(filename)
         self.load_state_dict(
-            T.load(os.path.join(self.chkpt_dir, filename),
+            T.load(os.path.join(filename),
                    map_location=self.device))

utils/metrics.py

@@ -25,6 +25,7 @@ def plot_avg_time(time_steps, num_players, fig_path):
     for player in time_steps:
         plt.plot(player)
     plt.savefig(os.path.join(fig_path, 'avg_time.png'))
+    plt.close()
 
 def plot_score(scores, num_players, figure_path):