exacity / deeplearningbook-chinese

Deep Learning Book Chinese Translation
https://exacity.github.io/deeplearningbook-chinese
35.61k stars 9.12k forks source link

Chapter5 #10

Closed liber145 closed 7 years ago

liber145 commented 7 years ago

关于第五章的建议,请放在这里。

huangpingchun commented 7 years ago

https://github.com/exacity/deeplearningbook-chinese/blob/master/Chapter5/machine_learning_basics.tex 145行 痛常 =》通常

liber145 commented 7 years ago

@huangpingchun 好的,多谢指出来

huangpingchun commented 7 years ago

estimator:chap5}的\firstgls{variance}就是一个方差 5.4.3 节, “估计的方差就是一个方差” 这句读着有点别扭?

huangpingchun commented 7 years ago

5.6 节 “prior belief” 一般概率书中都翻译为 “先验知识” 吧? 先验信念 好像没有专业书有这种翻译

liber145 commented 7 years ago

@huangpingchun 现在确实有点别扭。那个variance出现的地方,原句是 “The variance of an estimator is simply the variance”。暂时也没想到更好的翻译了。

huangpingchun commented 7 years ago

点积 和 内积 在5.7.2交替出现,后续全书应该采用统一的术语

liber145 commented 7 years ago

@huangpingchun 那个prior belief的翻译是从这来的,https://www.zhihu.com/question/20587681,里面将'prior belief‘翻译成“前置信念”。我觉得前置翻译成先验好一些。然后deep belief network里面的belief我们翻译成信念,所以这里也用信念了。

liber145 commented 7 years ago

@huangpingchun 点积 和 内积,分别对应着 dot product 和 inner product。我把它们都加到 terminology 中去吧。谢谢指出~

huangpingchun commented 7 years ago

好的,多谢释疑。

huangpingchun commented 7 years ago

“ @huangpingchun 现在确实有点别扭。那个variance出现的地方,原句是 “The variance of an estimator is simply the variance”。暂时也没想到更好的翻译了。 ” ----- 关于这一点 我再看了下原文, 还有一个疑问,你这个小节标题 “方差和标准差”,standard error 对应的翻译应该是“标准误” 而不是 “标准差”,它是 “样本统计量的标准差” https://zh.wikipedia.org/wiki/%E6%A0%87%E5%87%86%E8%AF%AF ,其数学符号 文中也是用的SE 而不是 SD(标准差)。是不是原作者 指代的 估计的方差(The variance of an estimator)其实是一种 “样本统计量”的方差,所以才说的它是一种简单的方差呢?一点不成熟的小疑问。

huangpingchun commented 7 years ago

https://zh.wikipedia.org/wiki/%E6%A0%87%E5%87%86%E8%AF%AF

liber145 commented 7 years ago

@huangpingchun 涨知识啦。多谢指出~

futianfan commented 7 years ago

@liber145 我们还是要学习一个,要不断的提高自己的知识水平。

huangpingchun commented 7 years ago

嘿嘿!大家一起学习,都希望这本书成为经典

SwordYork commented 7 years ago

多谢@huangpingchun 大神细心指点, @exacity/translation 全体对此表示非常感谢!

SwordYork commented 7 years ago

@liber145 我来不及改了,公式没有标点。5.1等字体不对,麻烦好好检查。

SwordYork commented 7 years ago

第一句话反了,我也没发现。。

huangpingchun commented 7 years ago

5.11.1 小节 (1717行) “这种现象被称为\firstgls{curse_of_dimensionality} 维数灾难(curse of dimensionality)。” 后面double了一次“维数灾难” 名词。

liber145 commented 7 years ago

@huangpingchun 嗯,看到了。谢谢指正!

liber145 commented 7 years ago

@SwordYork OK,给改过来了。

fairmiracle commented 7 years ago

1, https://github.com/exacity/deeplearningbook-chinese/blob/master/Chapter5/machine_learning_basics.tex#L248 $\Vy^{(j)}$==》$\Vx^{(j)}$

2, https://github.com/exacity/deeplearningbook-chinese/blob/master/Chapter5/machine_learning_basics.tex#L344 https://github.com/exacity/deeplearningbook-chinese/blob/master/Chapter5/machine_learning_basics.tex#L459 两处in action翻译成动态,有没有更好的说法。毕竟两个图都没反映动态特征

3,好几处More frequently, more formally等副词直译都比较硬,请再斟酌

4, https://github.com/exacity/deeplearningbook-chinese/blob/master/Chapter5/machine_learning_basics.tex#L1285 圆括号不成对,应该是\firstgls{logistic_regression}编译完了自动括号,句末括号成单了,不知道其它地方有没有

5, https://github.com/exacity/deeplearningbook-chinese/blob/master/Chapter5/machine_learning_basics.tex#L1620 \gls{minibatch}$\SetB$ 中间隔断一下,看着难受。。。

liber145 commented 7 years ago

@fairmiracle 谢谢帮忙校对! 1,4已定位改正。

2,这两处in action译作动态确实欠妥。现打算译作“在使用中的”,你看如何。 “For an example of the linear regression learning algorithm in action, see figure 5.1.” 参看图5.1,线性回归算法在使用中的示例。

Figure 5.2 shows this principle in action. 图5.2展示了这个原理在使用中的情况。

3, More frequently, the setting must be a hyperparameter because it is not appropriate to learn that hyperparameter on the training set. 更多的情况是,该设定必须是超参数,因为它不适合在训练集上学习。

a.We frequently use the term “linear” ... b. ... this arises frequently in the context of density estimation. c. ... should we penalize the system more if it frequently makes medium-sized mistakes ... d. This kind of situation arises frequently in medical diagnosis ... 我觉得abd用“经常”好,c用“频繁”好。“频繁”有点贬义。我经常散步 vs 我频繁散步

式5.37 和 式5.55,两处出现了more formally,后面接着公式,感觉译成“更形式地”可以的。“formally”,“informally”,还有其他副词,会再斟酌。

  1. \gls{minibatch}暂时还没给中文翻译,后期如果确定使用英文的话,会注意隔断下。

多谢啦~

huangpingchun commented 7 years ago

5.1.1 /* \item \textbf{缺失值归责}:这类任务中,\gls{ML}算法给定一个新\gls{example:chap5}$\Vx\in\SetR^n$,$\Vx$中某些元素$x_i$缺失。 算法必须预测这些缺失值。 ***/ "缺失值归责" ? =》 “缺失值填补(imputation of missing values)”。

huangpingchun commented 7 years ago

5.1.3 “尽管\gls{unsupervised_learning}和\gls{supervised_learning}并非完全正式或不同的概念” 这句话太拗口了。=》“尽管\gls{unsupervised_learning}和\gls{supervised_learning}并非十分严谨的概念”? 请斟酌!

huangpingchun commented 7 years ago

”Vapnik-Chervonenkis 维度“ 可以直接译为 “VC维”, 可参考 清华张学工老师 的两本译作 《统计学习理论的本质》和 《统计学习理论》

liber145 commented 7 years ago

@huangpingchun

  1. "imputation of missing data",有个矩阵补全算法叫 soft-impute,对应的还有个 hard-impute,两个的区别是一个用核范数做惩罚,一个用L0范数做惩罚。所以翻译时用了“归责”。这里结合文中具体说明的内容,确实“填补”或“补全”会更贴切些。如果文中用completion,就不用纠结这个问题了。额,我需要衡量下这个译法。或者有在什么地方看到 imputation of missing data 或 soft-impute 的中文翻译么

  2. 我觉得它这里想表达 监督学习 和 非监督学习 不是两个没有交集的概念。直接写“并非十分严谨”的话,感觉没有传达出这个意思;我那样翻译又很拗口。 翻译成“尽管\gls{unsupervised_learning}和\gls{supervised_learning}并非完全没有交集的正式概念”。 这样如何?

  3. 好的。保留第一次出现的全译,后面的修整为“VC维”。

多谢指正!

fairmiracle commented 7 years ago

imputation直接翻译成“填补”就好,completion是“补全”,“归责”太硬

SwordYork commented 7 years ago

@liber145 @fairmiracle @huangpingchun 看了下wiki,插补,填补都可以。补全不太好。

fairmiracle commented 7 years ago

@SwordYork 这个词在生物统计中很常用,因为gene expression data常有缺失,所以一般叫做缺失数据的填补,著名的impute包已经存在十几年了 http://bioconductor.org/packages/release/bioc/html/impute.html

huangpingchun commented 7 years ago

第一个问题: 我找到了 Statistical Analysis with Missing Data, Second Edition 这本书 对应的 中文翻译版本 《缺失数据统计分析(第二版)》 里面对缺失值做了很全的论述,imputation 基本上都是翻译为“借补” ,而 “填补” 书中用了 filling。

SwordYork commented 7 years ago

@liber145 @fairmiracle @huangpingchun 借补,插补,填补。我觉得我得问问我们导师,让他拍砖。多谢大家建议!

liber145 commented 7 years ago

@SwordYork @huangpingchun @fairmiracle 谢谢大家的反馈!我目前倾向于填补。我向导师反馈下,再商量下怎么好。

liber145 commented 7 years ago

@SwordYork @huangpingchun @fairmiracle 现在把 imputation 翻译成 填补 了。谢谢各位!

Doulrs commented 7 years ago

\firstall{SVM}是\gls{unsupervised_learning}中最有影响力的方法之一\citep{Boser92,Cortes95}

SVM 是 supervised learning。 原文中为One of the most influential approaches to supervised learning is the support vector machine

Doulrs commented 7 years ago

关于流形学习 manifold learning。 但是, Earth is Earth is a two-dimensional manifold or a surface in 3D space. 翻译应为三维空间中的二维流形或者二维流形。 PS: The space looks locally like the Euclidean plane, but it may have different global structure.

huangpingchun commented 7 years ago

无独有偶!5.7.1小节开始部分“本书的大部分无监督学习算法……” 。也错了,原文是监督学习。一并修改吧

liber145 commented 7 years ago

@Bojian @huangpingchun OK. 两处 无监督学习 已定位改正!多谢指正!

@Bojian “三维空间中的二维流形” 是说 “In everyday life, we experience the surface of the world as a 2-D plane, but it is in fact a spherical manifold in 3-D space.” 这句么? 我现在译作 “日常生活中,我们将地球视为二维平面,但实际上它是三维空间中的球状流形。” 你看如何。 Earth is a two-dimensional manifold 那句我没找到。。。

Doulrs commented 7 years ago

@liber145 那一句是以前在课本上看到的 :D. 你的翻译挺好的 会继续支持你们的

JoyFYan commented 7 years ago

错别字: 723行,“返回一个接近真实植”其中“植”改为“值”

liber145 commented 7 years ago

@JoyFYan 额,昨天刚看过这个地方,没有发现出来。。多谢指正!

huangpingchun commented 7 years ago

1317行 “在某些无穷维空间中,.......” 这一章别的地方infinite都翻译为 “无限”了,数学术语保持一致吧! 1333行 “在某些情况下,$\phi(\Vx)$甚至可以使无限维的......” 有个错别字 “使”。

liber145 commented 7 years ago

@huangpingchun OK. 我把infinite给核查一遍。谢谢指正!

zdx3578 commented 7 years ago

5.8.2的独热码 看着很费解,保留 one-hot 应该更方便理解,或至少保留对照的英文。

SwordYork commented 7 years ago

@zdx3578 多谢!应该根据gls来,这里@liber145 没来得及改。 已经改好了。

zdx3578 commented 7 years ago

感谢各位的辛勤付出!

SwordYork commented 7 years ago

@exacity/translation @zdx3578 要翻得更准确,需要大家一起努力。多谢鼓励!

HeimingX commented 7 years ago

第5.9小节,公式(5.98)上面:“重要的是,当训练集大小m′ 增长时,m 通常是固定的”中的两个符号m′和m颠倒了

liber145 commented 7 years ago

@HeimingX 好的,多谢指正!

liber145 commented 7 years ago

@fairmiracle @Bojian @zdx3578 @HeimingX 多谢各位帮忙校对!近来翻译组大致完成一期校对,二期校对希望网友加入帮忙勘误。不知大家近段时间有空没,能不能帮忙二期中英对比校对工作。

fairmiracle commented 7 years ago

@liber145 中英对比,第五章 机器学习基础 三月前完成