Fixed errors for MARL

agents/ppo/agent.py

@@ -15,28 +15,20 @@ class Agent:
         self.n_epochs = n_epochs
         self.gae_lambda = gae_lambda
 
-        print("Preparing Actor model...")
         self.actor = ActorNetwork(input_dims, n_actions, alpha)
-        print(f"Actor network activated using {self.actor.device}")
-        print("\nPreparing Critic model...")
         self.critic = CriticNetwork(input_dims, alpha)
-        print(f"Critic network activated using {self.critic.device}")
         self.memory = PPOMemory(batch_size)
 
     def remember(self, state, action, probs, vals, reward, done):
         self.memory.store_memory(state, action, probs, vals, reward, done)
 
-    def save_models(self):
-        print('... saving models ...')
-        self.actor.save_checkpoint()
-        self.critic.save_checkpoint()
-        print('... done ...')
+    def save_models(self, actr_chkpt = 'actor_ppo', crtc_chkpt = 'critic_ppo'):
+        self.actor.save_checkpoint(actr_chkpt)
+        self.critic.save_checkpoint(crtc_chkpt)
 
-    def load_models(self):
-        print('... loading models ...')
-        self.actor.load_checkpoint()
-        self.critic.load_checkpoint()
-        print('.. done ...')
+    def load_models(self, actr_chkpt = 'actor_ppo', crtc_chkpt = 'critic_ppo'):
+        self.actor.load_checkpoint(actr_chkpt)
+        self.critic.load_checkpoint(crtc_chkpt)
 
     def choose_action(self, observation):
         state = T.tensor(observation, dtype=T.float).to(self.actor.device)
@@ -56,7 +48,7 @@ class Agent:
             state_arr, action_arr, old_probs_arr, vals_arr, reward_arr, dones_arr, batches = self.memory.generate_batches()
 
             values = vals_arr
-            advantage = np.zeros(len(reward_arr), dtype=np.float32)
+            advantage = np.zeros(len(reward_arr), dtype=np.float64)
 
             for t in range(len(reward_arr)-1):
                 discount = 1

agents/ppo/brain.py

@@ -55,7 +55,7 @@ class ActorNetwork(nn.Module):
     def __init__(self, input_dim, output_dim, alpha, fc1_dims=256, fc2_dims=256, chkpt_dir='tmp/ppo'):
         super(ActorNetwork, self).__init__()
 
-        self.checkpoint_file = os.path.join(chkpt_dir, 'actor_torch_ppo')
+        self.chkpt_dir = chkpt_dir
 
         self.actor = nn.Sequential(
             nn.Linear(input_dim, fc1_dims),
@@ -68,8 +68,7 @@ class ActorNetwork(nn.Module):
 
         self.optimizer = optim.Adam(self.parameters(), lr=alpha, eps=1e-5)
 
-        self.device = T.device('cuda:0' if T.cuda.is_available() else (
-            'mps' if T.backends.mps.is_available() else 'cpu'))
+        self.device = T.device('cuda:0' if T.cuda.is_available() else 'cpu')
 
         self.to(self.device)
 
@@ -79,11 +78,11 @@ class ActorNetwork(nn.Module):
 
         return dist
 
-    def save_checkpoint(self):
-        T.save(self.state_dict(), self.checkpoint_file)
+    def save_checkpoint(self, filename = 'actor_torch_ppo'):
+        T.save(self.state_dict(), os.path.join(self.chkpt_dir, filename))
 
-    def load_checkpoint(self):
-        self.load_state_dict(T.load(self.checkpoint_file))
+    def load_checkpoint(self, filename = 'actor_torch_ppo'):
+        self.load_state_dict(T.load(os.path.join(self.chkpt_dir, filename)))
 
 
 class CriticNetwork(nn.Module):
@@ -91,7 +90,7 @@ class CriticNetwork(nn.Module):
     def __init__(self, input_dims, alpha, fc1_dims=256, fc2_dims=256, chkpt_dir='tmp/ppo'):
         super(CriticNetwork, self).__init__()
 
-        self.checkpoint_file = os.path.join(chkpt_dir, 'critic_torch_ppo')
+        self.chkpt_dir = chkpt_dir
 
         self.critic = nn.Sequential(
             nn.Linear(input_dims, fc1_dims),
@@ -103,8 +102,7 @@ class CriticNetwork(nn.Module):
 
         self.optimizer = optim.Adam(self.parameters(), lr=alpha, eps=1e-5)
 
-        self.device = T.device('cuda:0' if T.cuda.is_available() else (
-            'mps' if T.backends.mps.is_available() else 'cpu'))
+        self.device = T.device('cuda:0' if T.cuda.is_available() else 'cpu')
 
         self.to(self.device)
 
@@ -112,8 +110,8 @@ class CriticNetwork(nn.Module):
         value = self.critic(state)
         return value
 
-    def save_checkpoint(self):
-        T.save(self.state_dict(), self.checkpoint_file)
+    def save_checkpoint(self, filename = 'critic_torch_ppo'):
+        T.save(self.state_dict(), os.path.join(self.chkpt_dir, filename))
 
-    def load_checkpoint(self):
-        self.load_state_dict(T.load(self.checkpoint_file))
+    def load_checkpoint(self, filename = 'critic_torch_ppo'):
+        self.load_state_dict(T.load(os.path.join(self.chkpt_dir, filename)))

configs/game/monster_config.py

@@ -9,4 +9,4 @@ monster_data = {
     'squid': {'id': 1, 'health': .1, 'exp': 1, 'attack': .5, 'attack_type': 'slash', 'speed': 3, 'knockback': 20, 'attack_radius': 80, 'notice_radius': 360},
     'raccoon': {'id': 2, 'health': .3, 'exp': 2.5, 'attack': .8, 'attack_type': 'claw',  'speed': 2, 'knockback': 20, 'attack_radius': 120, 'notice_radius': 400},
     'spirit': {'id': 3, 'health': .1, 'exp': 1.1, 'attack': .6, 'attack_type': 'thunder', 'speed': 4, 'knockback': 20, 'attack_radius': 60, 'notice_radius': 350},
-    'bamboo': {'id': 4, 'health': .07, 'exp': 1.2, 'attack': .4, 'attack_type': 'leaf_attack', 'speed': 3, 'knockback': 20, 'attack_radius': 50, 'notice_radius': 300}}
+    'bamboo': {'id': 4, 'health': .07, 'exp': 1.2, 'attack': .2, 'attack_type': 'leaf_attack', 'speed': 3, 'knockback': 20, 'attack_radius': 50, 'notice_radius': 300}}

entities/player.py

@@ -120,9 +120,9 @@ class Player(pygame.sprite.Sprite):
 
         self.reward_features = [
             self.stats.exp,
-            np.exp(-nearest_dist**2),
-            np.exp(-nearest_enemy.stats.health**2),
-            -np.exp(-self.stats.health)
+            2*np.exp(-nearest_dist**2),
+            np.exp(-nearest_enemy.stats.health),
+            -np.exp(-self.stats.health**2)
         ]
 
         self.state_features = [
@@ -158,20 +158,21 @@ class Player(pygame.sprite.Sprite):
         self.num_features = len(self.state_features)
 
     def setup_agent(self):
+        print(f"Initializing Agent {self.player_id} ...")
         self.agent = Agent(
             input_dims=len(self.state_features),
             n_actions=len(self._input.possible_actions),
             batch_size=5,
             n_epochs=4)
+        print(f" Agent initialized using {self.agent.actor.device}. Attempting to load models ...")
         try:
-            self.agent.load_models()
+            self.agent.load_models(actr_chkpt = f"player_{self.player_id}_actor", crtc_chkpt = f"player_{self.player_id}_critic")
+            print("Models loaded ...\n")
         except FileNotFoundError:
-            print("FileNotFoundError for agent.load_model().\
-                Skipping loading...")
+            print("FileNotFound for agent. Skipping loading...\n")
 
     def is_dead(self):
         if self.stats.health <= 0:
-            self.stats.exp = max(0, self.stats.exp - .5)
             return True
         else:
             return False
@@ -202,9 +203,6 @@ class Player(pygame.sprite.Sprite):
 
         self.get_current_state()
 
-        if self.is_dead():
-            self.agent.learn()
-
         # Refresh objects based on input
         self.status = self._input.status
 
@@ -218,3 +216,6 @@ class Player(pygame.sprite.Sprite):
         self.stats.health_recovery()
         self.stats.energy_recovery()
         self._input.cooldowns(self._input.combat.vulnerable)
+
+        if self.is_dead():
+            self.stats.exp = max(-1, self.stats.exp - .5)

game.py

@@ -12,7 +12,7 @@ class Game:
         pygame.init()
 
         self.screen = pygame.display.set_mode(
-            (WIDTH, HEIGHT))
+            (WIDTH, HEIGHT), pygame.HIDDEN)
 
         pygame.display.set_caption('Pneuma')
 
@@ -22,8 +22,11 @@ class Game:
 
         self.level = Level()
 
+        self.max_num_players = len(self.level.player_sprites)
+
     def calc_score(self):
-        self.scores = [0 for _ in range(len(self.level.player_sprites))]
+
+        self.scores = [0 for _ in range(self.max_num_players)]
 
         for player in self.level.player_sprites:
             self.scores[player.player_id] = player.stats.exp
@@ -39,7 +42,7 @@ class Game:
 
         self.screen.fill(WATER_COLOR)
 
-        self.level.run(who='observer')
+        self.level.run()
 
         pygame.display.update()
         self.clock.tick(FPS)

level/level.py

@@ -172,17 +172,9 @@ class Level:
         if who == 'observer':
             self.visible_sprites.custom_draw(self.observer)
             self.ui.display(self.observer)
-        elif who == 'player':
-            self.visible_sprites.custom_draw(self.player)
-            self.ui.display(self.player)
 
         debug('v0.8')
 
-        for player in self.player_sprites:
-            if player.is_dead():
-                self.dead_players[player.player_id] = True
-
-        self.done = True if self.dead_players.all() == 1 else False
 
         if not self.game_paused:
             # Update the game
@@ -191,8 +183,16 @@ class Level:
 
             self.get_players_enemies()
             self.get_distance_direction()
-            self.apply_damage_to_player()
             self.visible_sprites.update()
+            self.apply_damage_to_player()
 
         else:
             debug('PAUSED')
+
+        for player in self.player_sprites:
+            if player.is_dead():
+                self.dead_players[player.player_id] = True
+
+        self.done = True if self.dead_players.all() == 1 else False
+
+

main.py

@@ -1,3 +1,5 @@
+import random
+import torch as T
 import numpy as np
 import matplotlib.pyplot as plt
 
@@ -8,23 +10,24 @@ from os import environ
 environ['PYGAME_HIDE_SUPPORT_PROMPT'] = '1'
 
 
-if __name__ == '__main__':
+random.seed(1)
+np.random.seed(1)
+T.manual_seed(1)
 
 n_episodes = 1000
 game_len = 10000
 
 figure_file = 'plots/score.png'
-    best_score = 0
-    avg_score = 0
 
 game = Game()
 
-    agent_list = []
-    exp_points_list = []
+agent_list = [0 for _ in range(game.max_num_players)]
+
 score_history = np.zeros(
-        shape=(len(game.level.player_sprites), n_episodes, ))
-    best_score = np.zeros(len(game.level.player_sprites))
-    avg_score = np.zeros(len(game.level.player_sprites))
+    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:
@@ -35,7 +38,7 @@ if __name__ == '__main__':
                 player.agent = agent_list[player.player_id]
                 player.stats.exp = score_history[player.player_id][i-1]
 
-        agent_list = []
+    agent_list = [0 for _ in range(game.max_num_players)]
 
     for j in range(game_len):
         if not game.level.done:
@@ -43,30 +46,35 @@ if __name__ == '__main__':
             game.run()
             game.calc_score()
 
-                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
+                if player.is_dead():
+                    agent_list[player.player_id] = 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
         else:
             break
 
     for player in game.level.player_sprites:
-            agent_list.append(player.agent)
+        if not player.is_dead():
+            agent_list[player.player_id] = 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 avg_score[player.player_id] >= best_score[player.player_id]:
-                player.agent.save_models()
+        if avg_score[player.player_id] > best_score[player.player_id]:
             best_score[player.player_id] = avg_score[player.player_id]
+            print(f"Saving models for agent {player.player_id}...")
+            player.agent.save_models(actr_chkpt = f"player_{player.player_id}_actor", crtc_chkpt = f"player_{player.player_id}_critic")
+            print("Models saved ...\n")
 
         print(
-                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]}")
+            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])