Fix relu inplace errors

.gitignore

@@ -142,3 +142,4 @@ dmypy.json
 rllib-env/
 runs/
 wandb/
+maps_old/

environment.py

@@ -101,6 +101,7 @@ def reward_done_info(self, reward, done, info):
         attack_bonus = 0
         gold_bonus = 0  # GoldBonus
         harvest_bonus = 0
+        give_bonus = 0
         level_bonus = 0
         if self.agent_id in self.env.realm.players:
             log = self.env.realm.event_log.get_data(agents=[self.agent_id],
@@ -119,16 +120,25 @@ def reward_done_info(self, reward, done, info):
             llog = self.env.realm.event_log.get_data(agents=[self.agent_id],
                                                      event_code=EventCode.LEVEL_UP,
                                                      tick=self.env.realm.tick)
-            hlog = self.env.realm.event_log.get_data(agents=[self.agent_id],
-                                                     event_code=EventCode.HARVEST_ITEM,
-                                                     tick=self.env.realm.tick)
-            # if llog.shape[0] > 0 and llog[0][-4] > 0:
-            #     # print('\033[93m' + 'level up log', log,
-            #     #       'agent_id', self.agent_id, '\033[0m')
-            #     level_bonus = 0.01
-            if hlog.shape[0] > 0 and hlog[0][-3] > 0:
-                # print('\033[94m' + 'harvest log', log,
+            harvest_log = self.env.realm.event_log.get_data(agents=[self.agent_id],
+                                                            event_code=EventCode.HARVEST_ITEM,
+                                                            tick=self.env.realm.tick)
+            give_log = self.env.realm.event_log.get_data(agents=[self.agent_id],
+                                                         event_code=EventCode.GIVE_ITEM,
+                                                         tick=self.env.realm.tick)
+            give_gold_log = self.env.realm.event_log.get_data(agents=[self.agent_id],
+                                                              event_code=EventCode.GIVE_GOLD,
+                                                              tick=self.env.realm.tick)
+            # Combat, Fishing  Skills
+            if llog.shape[0] > 0 and llog[0][-4] > 0 and (llog[0][-5] in range(1, 5)):
+                # print('\033[93m' + 'level up log', llog,
                 #       'agent_id', self.agent_id, '\033[0m')
+                level_bonus = 0.01
+            if give_log.shape[0] > 0 and give_log[0][-3] > 0:
+                give_bonus = 0.03
+            if give_gold_log.shape[0] > 0 and give_gold_log[0][-3] > 0:
+                give_bonus = 0.03
+            if harvest_log.shape[0] > 0 and harvest_log[0][-3] > 0:
                 harvest_bonus = 0.01
 
         # Add meandering bonus to encourage moving to various directions
@@ -149,7 +159,8 @@ def reward_done_info(self, reward, done, info):
         explore_bonus *= self.explore_bonus_weight
 
         reward = reward + explore_bonus + healing_bonus + \
-            meander_bonus + attack_bonus + gold_bonus + harvest_bonus+level_bonus
+            meander_bonus + attack_bonus + gold_bonus + \
+            harvest_bonus+level_bonus + give_bonus
 
         return reward, done, info
 

evaluate.py

@@ -158,6 +158,15 @@ def make_policy(envs):
     avg_completed_tick = np.mean([task.progress_info["completed_tick"]
                                   for task in nmmo_env.tasks if task.completed])
     print(f"Average completed tick: {avg_completed_tick:.1f}")
+    # can you save it inside csv with above 4 lines?
+    print("--------------------------------------------------")
+    df = pd.DataFrame({'Task': [nmmo_env.tasks[0].spec_name],
+                      'Number of agents completed the task': [num_completed],
+                       'Average maximum progress': [avg_progress],
+                       'Average completed tick': [avg_completed_tick]})
+    # df.to_csv(os.path.join('results', 'task.csv'), index=False)
+    with open(os.path.join('results', 'task.csv'), 'a') as f:
+        df.to_csv(f, header=False, index=False)
 
     # Save the replay file
     replay_file = os.path.join(

reinforcement_learning/config.py

@@ -13,7 +13,7 @@ class Config:
     # record_loss = False  # log all minibatch loss and actions, for debugging
 
     # Trainer Args
-    seed = 17
+    seed = 9
     num_cores = None  # Number of cores to use for training
     num_envs = 6  # Number of environments to use for training
     num_buffers = 2  # Number of buffers to use for training
@@ -23,7 +23,7 @@ class Config:
     eval_num_steps = 1_000_000  # 1_000_000  # Number of steps to evaluate
     checkpoint_interval = 5_000_000  # Interval to save models
     # f"nmmo_{time.strftime('%Y%m%d_%H%M%S')}_{seed}"  # Run name
-    run_name = f"test_nmmo_all_attn_hv_bonus_seed{seed}_exp15"
+    run_name = f"nmmo_rp_cr_attn_skill_bonus_seed{seed}_exp17"
     runs_dir = "./runs"  # Directory for runs
     policy_store_dir = None  # Policy store directory
     use_serial_vecenv = False  # Use serial vecenv implementation