Update rollout and undo neural baseline changes

neural-mmo

-../../neuralmmo_master/neural-mmo
\ No newline at end of file

players.yaml

@@ -4,9 +4,9 @@
 #  Number of opponent agents is exactly 127
 
 player_agent:
-  file: scripted_baseline_agent
-  agent_class: BaselineCombatAgent
-  agent_type: scripted
+  file: neural_baseline_agent
+  agent_class: NeuralBaselineAgent
+  agent_type: neural
 
 opponent_agents:
   agent_1:

resource

-../../neuralmmo_master/neural-mmo/resource
\ No newline at end of file

rollout.py

@@ -52,7 +52,7 @@ def run_episode(player_index, agents, N_TIME_STEPS):
 
         obs,dones,rewards,_ = env.step(actions,omitDead=True,preprocess=neural_agents)
 
-        for entid in alive_agents:
+        for entid in sorted(alive_agents):
             if entid not in list(obs.keys()):
                 dead_agents.append(entid)
 
@@ -64,26 +64,27 @@ def run_episode(player_index, agents, N_TIME_STEPS):
             actions[entid] = agents[entid_agent_map[entid]].compute_action(obs[entid])
         n_steps += 1
     
-    for entid in obs:
+    for entid in sorted(list(obs.keys())):
         if entid not in dead_agents:
             dead_agents.append(entid)
 
-    logs = env.terminal()    
-    player_entid = entid_agent_map[player_index]
+    logs = env.terminal()  
+    player_entid = agent_entid_map[player_index]  
+
     player_log = {}
-    player_log["Achievement"] = logs['Stats']['Achievement'][player_entid]
-    player_log["Equipment"] = logs['Stats']['Equipment'][player_entid]
-    player_log["Equipment"] = logs['Stats']['Exploration'][player_entid]
-    player_log["PlayerKills"] = logs['Stats']['PlayerKills'][player_entid]
-    player_log["Foraging"] = logs['Stats']['Foraging'][player_entid]
+    player_log["Achievement"] = logs['Stats']['Achievement'][dead_agents.index(player_entid)]
+    player_log["Equipment"] = logs['Stats']['Equipment'][dead_agents.index(player_entid)]
+    player_log["Exploration"] = logs['Stats']['Exploration'][dead_agents.index(player_entid)]
+    player_log["PlayerKills"] = logs['Stats']['PlayerKills'][dead_agents.index(player_entid)]
+    player_log["Foraging"] = logs['Stats']['Foraging'][dead_agents.index(player_entid)]
     return player_log
 
 
 def print_statistics(player_statistics,episode):
-    print("======= Episode {} ========".format(episode+1))
+    print("======= Episode {} =======".format(episode+1))
     print("Achievement ",player_statistics['Achievement'])
     print("Equipment ",player_statistics['Equipment'])
-    print("Equipment ",player_statistics['Equipment'])
+    print("Exploration ",player_statistics['Exploration'])
     print("PlayerKills ",player_statistics['PlayerKills'])
     print("Foraging ",player_statistics['Foraging'])
     print("=========================")
@@ -91,7 +92,7 @@ def print_statistics(player_statistics,episode):
 if __name__== "__main__":
     player_agent, opponent_agents = load_agents("players.yaml")
     N_EPISODES = 10
-    N_TIME_STEPS = 1024
+    N_TIME_STEPS = 102
     for episode in range(N_EPISODES):
         agents,player_index = assign_agents(player_agent,opponent_agents)
         statistics = run_episode(player_index,agents,N_TIME_STEPS)

rollout_update.py

-import sys
-sys.path.append('neural-mmo/')
-
-from forge.ethyr.torch import utils
-from forge.trinity.env import Env
-import projekt
-
-from utils.helpers import load_agents
-import random
-import copy
-
-def assign_agents(player_agent,opponent_agents):
-    player_index = 0 
-    if len(opponent_agents) != 127:
-        raise Exception("Number of opponent agents should add up to exactly 127")
-    random.shuffle(opponent_agents)
-    player_index = random.randint(0,127)
-    agents = copy.deepcopy(opponent_agents)
-    agents.insert(player_index,player_agent)
-    return agents,player_index
-
-
-def run_episode(player_index, agents, N_TIME_STEPS):
-    config = projekt.config.CompetitionRound1()
-    env = Env(config)
-    n_steps = 0
-    neural_agents = set()
-    dead_agents = []
-
-    obs = env.reset()
-    entids = list(obs.keys())
-    agent_entid_map = dict(zip(range(len(agents)), entids))
-    entid_agent_map = {x[1]:x[0] for x in agent_entid_map.items()}
-    for idx,agent in enumerate(agents):
-        if agent.type == 'neural':
-            neural_agents.add(agent_entid_map[idx])
-        
-    actions = {}
-    for entid in entids:
-        actions[entid] = agents[entid_agent_map[entid]].register_reset(obs[entid])
-
-    while len(obs.keys()) > 0 and n_steps < N_TIME_STEPS:
-        obs,dones,rewards,_ = env.step(actions,omitDead=False)
-        alive_agents = list(obs.keys())
-        for entid in dones:
-            if dones[entid]:
-                dead_agents.append(entid)
-        actions = {}
-        for entid in alive_agents:
-            if entid not in entid_agent_map:
-                continue
-            actions[entid] = agents[entid_agent_map[entid]].compute_action(obs[entid])
-        n_steps += 1
-    
-    for entid in obs:
-        if entid not in dead_agents:
-            dead_agents.append(entid)
-
-    logs = env.terminal()    
-    player_entid = entid_agent_map[player_index]
-    player_log = {}
-    if player_entid in dead_agents:
-        player_log["Achievement"] = logs['Stats']['Achievement'][player_entid]
-        player_log["Equipment"] = logs['Stats']['Equipment'][player_entid]
-        player_log["Exploration"] = logs['Stats']['Exploration'][player_entid]
-        player_log["PlayerKills"] = logs['Stats']['PlayerKills'][player_entid]
-        player_log["Foraging"] = logs['Stats']['Foraging'][player_entid]
-    return player_log
-
-
-def print_statistics(player_statistics,episode):
-    print(player_statistics, episode)
-
-if __name__== "__main__":
-    player_agent, opponent_agents = load_agents("players.yaml")
-    N_EPISODES = 10
-    N_TIME_STEPS = 10
-    for episode in range(N_EPISODES):
-        agents,player_index = assign_agents(player_agent,opponent_agents)
-        statistics = run_episode(player_index,agents,N_TIME_STEPS)
-        print_statistics(statistics,episode)