# training loop
while True:
current_state = agent.get_state()
next_action = agent.get_action(current_state)
game_over, score, reward = game.play_step(next_action)
new_state = agent.get_state()
agent.train_over_sample(current_state, next_action, reward, new_state,
game_over)
agent.save_values(current_state, next_action, reward, new_state,
game_over)
if game_over:
game.restart()
agent.num_of_games += 1
agent.train_over_batch()
if score > maximum_score:
# save checkpoint
maximum_score = score
agent.train.save_checkpoint()
print('Game', agent.num_of_games, 'Score', score, 'Record:',
maximum_score)
scores.append(score)
total_score += score
mean_score = total_score / agent.num_of_games
