Create ppo_vizdoom_v2.py

mains/ppo_vizdoom_v2.py

@@ -0,0 +1,58 @@
+import os,sys,inspect
+current_dir = os.path.dirname(os.path.abspath(inspect.getfile(inspect.currentframe())))
+parent_dir = os.path.dirname(current_dir)
+sys.path.insert(0, parent_dir) 
+
+import gym
+import tensorflow as tf
+
+import numpy as np
+from agents.PPO_v2 import Agent
+from environments.VizDoomEnv import VizDoomEnv
+from extras.statistics import Logger
+
+env = VizDoomEnv(scenario='defend_the_center.cfg')
+agent = Agent(n_actions=env.action_size,conv=True)
+L = Logger(name='log_ppo_viz')
+
+TEST_EPOCHS = 5
+PPO_STEPS = 256
+TARGET_SCORE = 200
+
+train_epochs = 0
+early_stop = False
+
+scores = []
+while not early_stop:
+    observation = env.reset()
+    score = 0
+    for _ in range(agent.PPO_STEPS):
+        action, log_probs, value = agent.choose_action(observation)
+        observation_, reward, done, info = env.step(action)
+        agent.store_experience(np.expand_dims(observation,axis=0),action,reward,np.expand_dims(observation_,axis=0),done,log_probs,value)
+        score += reward
+        if done:
+            observation = env.reset()
+            scores.append(score)
+            avg_score = np.mean(scores[-100:])
+            print('Episode:',len(scores),'Score:',score,'Avg score:',avg_score)
+            L.add_log('score',score)
+            L.add_log('avg_score',avg_score)
+            score = 0
+            if avg_score >= 195:
+                early_stop = True
+                break
+            continue
+        observation = observation_
+
+    obs = tf.convert_to_tensor([observation_])
+    next_value = agent.critic(obs)
+    next_value = next_value.numpy()[0][0]
+    states,actions,rewards,states_,dones,log_probs,values = agent.read_memory()
+    returns = agent.compute_gae(next_value,values,rewards,dones)
+    advantages = returns - values
+    agent.ppo_update(states,actions,log_probs,returns,advantages)
+
+if train_epochs%20==0:
+    L.save_game()
+train_epochs += 1