Implemented different sp and mp files

multi-agent.py

@@ -14,10 +14,10 @@ random.seed(1)
 np.random.seed(1)
 T.manual_seed(1)
 
-n_episodes = 10000
+n_episodes = 2000
 game_len = 5000
 
-figure_file = 'plots/score.png'
+figure_file = 'plots/scores_mp.png'
 
 game = Game()
 
@@ -34,9 +34,10 @@ for i in tqdm(range(n_episodes)):
         game.level.__init__(reset=True)
     # TODO: Make game.level.reset_map() so we don't pull out and load the agent every time (There is -definitevly- a better way)
         for player in game.level.player_sprites:
+
+            player.stats.exp = score_history[player.player_id][i-1]
             for agent in agent_list:
                 player.agent = agent_list[player.player_id]
-                player.stats.exp = score_history[player.player_id][i-1]
 
     agent_list = [0 for _ in range(game.max_num_players)]
 
@@ -55,8 +56,6 @@ for i in tqdm(range(n_episodes)):
             #     for player in game.level.player_sprites:
             #         for enemy in game.level.enemy_sprites:
             #             player.stats.exp *= .95
-        else:
-            break
 
     for player in game.level.player_sprites:
         if not player.is_dead():
@@ -76,7 +75,8 @@ for i in tqdm(range(n_episodes)):
             f"\nCumulative score for player {player.player_id}: {score_history[0][i]}\nAverage score for player {player.player_id}: {avg_score[player.player_id]}\nBest score for player {player.player_id}: {best_score[player.player_id]}")
 
 
-plt.plot(score_history[0])
+plt.plot(score_history)
+plt.savefig(figure_file)
 
 game.quit()
 

single-agent.py

@@ -0,0 +1,80 @@
+import random
+import torch as T
+import numpy as np
+import matplotlib.pyplot as plt
+
+from game import Game
+from tqdm import tqdm
+
+from os import environ
+environ['PYGAME_HIDE_SUPPORT_PROMPT'] = '1'
+
+
+random.seed(1)
+np.random.seed(1)
+T.manual_seed(1)
+
+n_episodes = 2000
+game_len = 5000
+
+figure_file = 'plots/score_sp.png'
+
+game = Game()
+
+agent_list = []
+
+score_history = np.zeros(shape=(game.max_num_players, n_episodes))
+best_score = np.zeros(game.max_num_players)
+avg_score = np.zeros(game.max_num_players)
+
+for i in tqdm(range(n_episodes)):
+    # TODO: Make game.level.reset_map() so we don't __init__ everything all the time (such a waste)
+    if i != 0:
+        game.level.__init__(reset=True)
+    # TODO: Make game.level.reset_map() so we don't pull out and load the agent every time (There is -definitevly- a better way)
+        for player in game.level.player_sprites:
+
+            player.stats.exp = score_history[player.player_id][i-1]
+            player.agent = agent_list[0]
+
+    agent_list = []
+
+    for j in range(game_len):
+        if not game.level.done:
+
+            game.run()
+            game.calc_score()
+
+            for player in game.level.player_sprites:
+                if player.is_dead():
+                    agent_list[0] = player.agent
+                    player.kill()
+
+            # if (j == game_len-1 or game.level.done) and game.level.enemy_sprites != []:
+            #     for player in game.level.player_sprites:
+            #         for enemy in game.level.enemy_sprites:
+            #             player.stats.exp *= .95
+
+    for player in game.level.player_sprites:
+        if not player.is_dead():
+            agent_list[0] = player.agent
+        exp_points = player.stats.exp
+        score_history[player.player_id][i] = exp_points
+        avg_score[player.player_id] = np.mean(
+            score_history[player.player_id])
+    if np.mean(avg_score) > np.mean(best_score):
+        best_score = avg_score
+        print("Saving models for agent...")
+        player.agent.save_models(
+            actr_chkpt="player_actor", crtc_chkpt="player_critic")
+        print("Models saved ...\n")
+
+    print(
+        f"\nAverage score: {np.mean(avg_score)}\nBest score: {np.mean(best_score)}")
+
+
+plt.plot(score_history)
+plt.savefig(figure_file)
+game.quit()
+
+plt.show()