Updated metrics

main.py

@@ -101,6 +101,12 @@ def main():
         episode_total_loss = np.zeros(
             shape=(n_agents, learnings_per_episode))
 
+        episode_entropy = np.zeros(
+            shape=(n_agents, learnings_per_episode))
+
+        episode_advantage = np.zeros(
+            shape=(n_agents, learnings_per_episode))
+
         # Main game loop
         for step in tqdm(range(episode_length),
                          leave=False,
@@ -127,6 +133,12 @@ def main():
                         episode_total_loss[player.player_id][learn_iters % learnings_per_episode]\
                             = player.agent.total_loss
 
+                        episode_entropy[player.player_id][learn_iters % learnings_per_episode]\
+                            = player.agent.entropy
+
+                        episode_advantage[player.player_id][learn_iters % learnings_per_episode]\
+                            = player.agent.advantage
+                        
                         learn_iters += 1
 
 
@@ -151,6 +163,12 @@ def main():
 
             time_alive[player.player_id][episode] = step
 
+            entropy[player.player_id][episode] = np.mean(
+                episode_entropy)
+
+            advantage[player.player_id][episode] = np.mean(
+                episode_advantage)
+
             # Check for new best score
             if score > best_score[player.player_id]:
                 print(f"\nEpisode:\

utils/metrics.py

@@ -3,7 +3,7 @@ import numpy as np
 import matplotlib.pyplot as plt
 
 
-def plot_learning_curve(scores, num_players, figure_path):
+def plot_learning_curve(scores, num_players, figure_path, ep_lenght):
 
     plt.figure()
     plt.title("Running Average - Score")
@@ -13,7 +13,7 @@ def plot_learning_curve(scores, num_players, figure_path):
     for score in scores:
         running_avg = np.zeros(len(score))
         for i in range(len(score)):
-            running_avg[i] = np.mean(score[max(0, i-100):(i+1)])
+            running_avg[i] = np.mean(score[max(0, i-int(ep_length/10)):(i+1)])
         plt.plot(running_avg)
     plt.savefig(os.path.join(figure_path, "avg_score.png"))
     plt.close()
@@ -40,7 +40,7 @@ def plot_score(scores, num_players, figure_path):
     plt.close()
 
 
-def plot_loss(nn_type, losses, num_players, figure_path):
+def plot_loss(nn_type, losses, num_players, figure_path, ep_length):
 
     plt.figure()
     plt.title(f"Running Average - {nn_type.capitalize()} Loss")
@@ -50,13 +50,13 @@ def plot_loss(nn_type, losses, num_players, figure_path):
     for loss in losses:
         running_avg = np.zeros(len(loss))
         for i in range(len(loss)):
-            running_avg[i] = np.mean(loss[max(0, i-100):(i+1)])
+            running_avg[i] = np.mean(loss[max(0, i-int(ep_length/10)):(i+1)])
         plt.plot(running_avg)
     plt.savefig(os.path.join(figure_path, f"{nn_type}_loss.png"))
     plt.close()
 
 
-def plot_parameter(name, parameter, num_players, figure_path):
+def plot_parameter(name, parameter, num_players, figure_path, ep_length):
 
     plt.figure()
     plt.title(f"Running Average - {name.capitalize()}")
@@ -66,7 +66,7 @@ def plot_parameter(name, parameter, num_players, figure_path):
     for param in parameter:
         running_avg = np.zeros(len(param))
         for i in range(len(param)):
-            running_avg[i] = np.mean(param[max(0, i-100):(i+1)])
+            running_avg[i] = np.mean(param[max(0, i-int(ep_length/10)):(i+1)])
         plt.plot(running_avg)
     plt.savefig(os.path.join(figure_path, f"{name}.png"))
     plt.close()