sess.run(tf.initialize_all_variables()) summary_writer = tf.train.SummaryWriter(args.save_folder, sess.graph_def) if args.restore != "No": saver.restore(sess, args.save_folder+args.restore) #just display games sim.reset() previous_state = numpy.zeros((sim.image_size,sim.image_size,3)) previous_state[:,:,0] = numpy.reshape(sim.screen,(sim.image_size,sim.image_size)) screen = sim.screen for i in range(1000): cv2.imshow('Phong!',cv2.resize(screen,(0,0),fx=2,fy=2)) key = cv2.waitKey(8) if key == UP_KEY: screen,score,points_made,end = sim.do_action(1,side="left") elif key == DOWN_KEY: screen,score,points_made,end = sim.do_action(2,side="left") else: screen,score,points_made,end = sim.do_action(0,side="left") action = learner.return_action(previous_state) screen,score,points_made,end = sim.do_action(action) previous_state[:,:,1:] = numpy.copy(previous_state[:,:,:2]) previous_state[:,:,0] = numpy.reshape(screen,(sim.image_size,sim.image_size)) previous_state_image = numpy.copy(previous_state) previous_state_image -= previous_state_image.min() previous_state_image /= previous_state_image.max() previous_state_image *= 255 cv2.imwrite('phong_state.png',previous_state_image) print(score)
sess.graph_def) if args.restore != "No": saver.restore(sess, args.save_folder + args.restore) #just display games sim.reset() previous_state = numpy.zeros((sim.image_size, sim.image_size, 3)) previous_state[:, :, 0] = numpy.reshape(sim.screen, (sim.image_size, sim.image_size)) screen = sim.screen for i in range(1000): cv2.imshow('Phong!', cv2.resize(screen, (0, 0), fx=2, fy=2)) key = cv2.waitKey(8) if key == UP_KEY: screen, score, points_made, end = sim.do_action( 1, side="left") elif key == DOWN_KEY: screen, score, points_made, end = sim.do_action( 2, side="left") else: screen, score, points_made, end = sim.do_action( 0, side="left") action = learner.return_action(previous_state) screen, score, points_made, end = sim.do_action(action) previous_state[:, :, 1:] = numpy.copy(previous_state[:, :, :2]) previous_state[:, :, 0] = numpy.reshape( screen, (sim.image_size, sim.image_size)) previous_state_image = numpy.copy(previous_state) previous_state_image -= previous_state_image.min() previous_state_image /= previous_state_image.max() previous_state_image *= 255