feat: Agent can now load model

#102

sc2_collectables_trainer.py

@@ -37,6 +37,7 @@ def declare_trainer():
 def main(unused_argv):
     try:
         trainer = declare_trainer()
+        #trainer.load("saves/")
         trainer.train()
         # trainer.play()
 

urnai/agents/agent_base.py

@@ -56,10 +56,4 @@ def learn(self, obs, reward, done) -> None:
         if self.previous_state is not None:
             next_state = self.state_space.update(obs)
             self.model.learn(self.previous_state, self.previous_action,
-                              reward, next_state, done)
-
-    def save(self, savepath) -> None:
-        if (self.model.persistence == None):
-            raise 'No persistence set in model'
-
-        self.model.persistence.save(savepath)
+                              reward, next_state, done)

urnai/models/dqn_pytorch.py

@@ -118,6 +118,7 @@ def __init__(self, action_wrapper: ActionSpaceBase, state_builder: StateBase,
             field_names=['state', 'action', 'reward', 'next_state', 'done'])
 
         self.persistence = PersistencePickle(self)
+        self.persistence.attr_block_list.append(['persistence'])
 
     def make_model(self):
         model = QNetwork(

urnai/models/model_base.py

@@ -29,4 +29,10 @@ def choose_action(self, state, excluded_actions, is_testing=False) -> int:
 		exploration algorithm for when is_testing=False.
 		One such exploration algorithm commonly used it the epsilon greedy strategy.
 		"""
-		pass
+		pass
+
+	def save(self, persist_path) -> None:
+		self.persistence.save(persist_path)
+
+	def load(self, persist_path) -> None:
+		self.persistence.load(persist_path)

urnai/trainers/trainer.py

@@ -25,6 +25,9 @@ def __init__(self, env, agent, max_training_episodes, max_playing_episodes, max_
     def train(self, reward_from_agent=True):
         self.training_loop(is_training=True, reward_from_agent=reward_from_agent)
 
+    def load(self, persist_path):
+        self.agent.model.load(persist_path)
+
     def play(self, reward_from_agent=True):
         self.training_loop(is_training=False, reward_from_agent=reward_from_agent)
 
@@ -88,7 +91,7 @@ def training_loop(self, is_training, reward_from_agent=True):
 
                 if done:
                     print("Episode: %d, Reward: %d" % (current_episodes, ep_reward))
-                    self.agent.save("saves/")
+                    self.agent.model.save("saves/")
                     break
 
             # if this is not a test (evaluation), saving is enabled and we are in a multiple