def run(ep,train=False):
pygame.init()
loss=[]
agent = DQN(3, 5)
env=pongGame()
weights_filepath = 'PongGame.h5'
if train==False:
agent.model.load_weights(weights_filepath)
print("weights loaded")
for e in range(ep):
for event in pygame.event.get():
if event.type == pygame.QUIT:
pygame.quit()
quit()
state = env.reset()
state = np.reshape(state, (1, 5))
score = 0
max_steps = 1000
for i in range(max_steps):
action = agent.act(state)
reward, next_state, done = env.step(action)
score += reward
next_state = np.reshape(next_state, (1, 5))
agent.remember(state, action, reward, next_state, done)
state = next_state
if train==True:
agent.replay()
if done:
print("episode: {}/{}, score: {}".format(e, ep, score))
break
loss.append(score)
if train:
agent.model.save_weights("PongGame.h5")
return loss
action = agent.act(pre_ob, step=i)
ob, reward, done, _ = env.step(action)
if reward <= -1:
reward = -1
next_pre_ob = preprocess(ob)
# Stack observations
next_pre_ob = next_pre_ob.reshape(1, 100, 100)
ob_stack = np.insert(ob_stack, -1, next_pre_ob, axis=3)
ob_stack = np.delete(ob_stack, 0, axis=3)
next_pre_ob = ob_stack
agent.remember(pre_ob, action, reward, next_pre_ob, done)
agent.replay()
pre_ob = next_pre_ob
score = score + reward
if done:
break
scores.append(score)
print("Episode {} score: {}".format(i + 1, score))
mean_score = np.mean(scores)
if (i + 1) % 5 == 0:
print(
"Episode {}, score: {}, exploration at {}%, mean of last 100 episodes was {}"
.format(i + 1, score, agent.epsilon * 100, mean_score))
