39%的准确性可能不那么令人印象深刻。但是,用这么少的数据研究18类分类问题也是很困难的,况且我们的模型相比零规则基线( Zero Rule Baseline)获得了20个百分点的增益,它是为了猜测所有精灵宝可梦中最常见的类。表1列出了测试集上每类出现的频率,这为零规则提供了19.5%的精度。
表1:键入测试数据集的频率。 但是,如果我们预期机器有一天能成为这个星球的统治者,我们就不应该用这种笨拙的方法衡量它们。距离计算机有一天在「精灵宝可梦分类挑战」中打败我的小兄弟,还有很长的路要走。但往好的地方想,他们可能已经击败了我爹,但这是另一篇文章的主题啦。 参考文献 Bulbapedia. (2017) Type. Available from: bulbapedia.bulbagarden.net/wiki/Type (Date of access: 20/01/2017). Go Game Guru. (2017) DeepMind AlphaGo vs Lee Sedol. Available from: https://gogameguru. com/tag/deepmind-alphago-lee-sedol/ (Date of access: 07/Mar/2017). Large Scale Visual Recognition Challenge.(2015) Large Scale Visual Recognition Challenge 2015 (ILSVRC2015). Available from: (Date of access: 20/01/2017). Scikit-Image. (2017) Module: filters. Available from: lters.html#skimage.filters.sobel (Date of access: 07/Mar/2017). Tromp, J. & Farnebäck, G.(2016) Combinatorics of Go. Available from: https://tromp.github.io/go/ gostate.pdf Veekun.(2017) Sprite Packs. Available from: https:// veekun.com/dex/downloads (Date of access: 20/01/2017). Wikipedia. (2017a) Artificial Neural Network. Available from: https://en.wikipedia.org/wiki/ Artificial_neural_network (Date of access: 07/Mar/2017). Wikipedia. (2017b) Pokémon. Available from: https://en.wikipedia.org/wiki/Pok%C3%A9mon (Date of access: 20/01/2017). 原文地址:https://jgeekstudies.files.wordpress.com/2017/03/soares_2017_neural-mon.pdf ©本文为机器之心编译,转载请联系本公众号获得授权。 ?------------------------------------------------ 加入机器之心(全职记者/实习生):[email protected] 投稿或寻求报道:[email protected] 广告&商务合作:[email protected] (责任编辑:本港台直播) |