if done: print("Epoch %d Train Game %d get %.1f" % (epoch, games_cnt, game.get_total_reward())) break if SAVE_MODEL and games_cnt % 100 == 0: saver.save(SESSION, model_savefile) print("Saving the network weigths to:", model_savefile) print("\nTesting...") test_scores = [] for test_step in range(EPISODES_TO_TEST): game.reset() agent.reset_cell_state() while not game.is_episode_finished(): state = game.get_state() action = agent.act(state, train=False) game.make_action(action) test_scores.append(game.get_total_reward()) test_scores = np.array(test_scores) if SAVE_MODEL: saveScore(test_scores) if test_scores.mean() > max_avgR: max_avgR = test_scores.mean() saver.save(SESSION, max_model_savefile) ''' print("TIME TO WATCH!!") # Reinitialize the game with window visible