统计学习基础

2009-1  

世界图书出版公司  

哈斯蒂  

533  

无  

The field of Statistics is constantly challenged by the problems that science and industry brings to its door. In the early days, these problems often came from agricultural and industrial experiments and were relatively small in scope. With the advent of computers and the information age, statistical problems have exploded both in size and complexity. Challenges in the areas of data storage, organization and searching have led to the new field of "data mining"; statistical and computational problems in biology and medicine have created "bioinformatics." Vast amounts of data are being generated in many fields, and the statistician's job is to make sense of it all: to extract important patterns and trends, and understand "what the data says." We call this learning from data.The challenges in learning from data have led to a revolution in the statistical sciences. Since computation plays such a key role, it is not surprising that much of this new development has been done by researchers in other fields such as computer science and engineering.The learning problems that we consider can be roughly categorized as either supervised or unsupervised. In supervised learning, the goal is to predict the value of an outcome measure based on a number of input measures; in unsupervised learning, there is no outcome measure, and the goal is to describe the associations and patterns among a set of input measures.

　　The learning problems that we consider can be roughly categorized as either supervised or unsupervised. In supervised learning, the goal is to predict the value of an outcome measure based on a number of input measures; in unsupervised learning, there is no outcome measure, and the goal is to describe the associations and patterns among a set of input measures.

作者：(美国)哈斯蒂 (Hastie.T.)

Preface1 Introduction Overview of Supervised Learning2.1 Introduction2.2 Variable Types and Terminology2.3 Two Simple Approaches to Prediction: Least Squares and Nearest Neighbors2.3.1 Linear Models and Least Squares2.3.2 Nearest-Neighbor Methods2.3.3 From Least Squares to Nearest Neighbors2.4 Statistical Decision Theory2.5 Local Methods in High Dimensions2.6 Statistical Models, Supervised Learning and Function Approximation2.6.1 A Statistical Model for the Joint Distribution Pr(X,Y)2.6.2 Supervised Learning2.6.3 Function Approximation2.7 Structured Regression Models2.7.1 Difficulty of the Problem2.8 Classes of Restricted Estimators2.8.1 Roughness Penalty and Bayesian Methods2.8.2 Kernel Methods and Local Regression2.8.3 Basis Functions and Dictionary Methods2.9 Model Selection and the Bias-Variance TradeoffBibliographic Notes Exercises 3 Linear Methods for Regression3.1 Introduction3.2 Linear Regression Models and Least Squares 3.2.1 Example：Prostate Cancer3.2.2 The Ganss-Markov Theorem3.3 Multiple Regression from Simple Univariate Regression3.3.1 Multiple Outputs3.4 Subset Selection and Coefficient Shrinkage3.4.1 Subset Selection3.4.2 Prostate Cancer Data Example fContinued)3.4.3 Shrinkage Methods3.4.4 Methods Using Derived Input Directions3.4.5 Discussion：A Comparison of the Selection and Shrinkage Methods3.4.6 Multiple Outcome Shrinkage and Selection 3.5 Compntational ConsiderationsBibliographic NotesExercises 4 Linear Methods for Classification4.1 Introduction4.2 Linear Regression of an Indicator Matrix4.3 Linear Discriminant Analysis4.3.1 Regularized Discriminant Analysis4.3.2 Computations for LDA 4.3.3 Reduced-Rank Linear Discriminant Analysis 4.4 Logistic Regression4.4.1 Fitting Logistic Regression Models4.4.2 Example：South African Heart Disease4.4.3 Quadratic Approximations and Inference4.4.4 Logistic Regression or LDA74.5 Separating Hyper planes4.5.1 Rosenblatt's Perceptron Learning Algorithm4.5.2 Optimal Separating Hyper planesBibliographic NotesExercises 5 Basis Expansions and Regularizatlon5.1 Introduction5.2 Piecewise Polynomials and Splines5.2.1 Natural Cubic Splines5.2.2 Example: South African Heart Disease (Continued) 5.2.3 Example: Phoneme Recognition5.3 Filtering and Feature Extraction5.4 Smoothing Splines5.4.1 Degrees of Freedom and Smoother Matrices5.5 Automatic Selection of the Smoothing Parameters5.5.1 Fixing the Degrees of Freedom5.5.2 The Bias-Variance Tradeoff5.6 Nonparametric Logistic Regression5.7 Multidimensional Splines5.8 Regularization and Reproducing Kernel Hilbert Spaces . . 5.8.1 Spaces of Phnctions Generated by Kernels5.8.2 Examples of RKHS5.9 Wavelet Smoothing5.9.1 Wavelet Bases and the Wavelet Transform5.9.2 Adaptive Wavelet FilteringBibliographic NotesExercisesAppendix: Computational Considerations for SplinesAppendix: B-splinesAppendix: Computations for Smoothing Splines6 Kernel Methods6.1 One-Dimensional Kernel Smoothers6.1.1 Local Linear Regression6.1.2 Local Polynomial Regression6.2 Selecting the Width of the Kernel6.3 Local Regression in Jap6.4 Structured Local Regression Models in ]ap6.4.1 Structured Kernels6.4.2 Structured Regression Functions6.5 Local Likelihood and Other Models6.6 Kernel Density Estimation and Classification6.6.1 Kernel Density Estimation6.6.2 Kernel Density Classification6.6.3 The Naive Bayes Classifier6.7 Radial Basis Functions and Kernels6.8 Mixture Models for Density Estimation and Classification 6.9 Computational ConsiderationsBibliographic Notes Exercises7 Model Assessment and Selection7.1 Introduction7.2 Bias, Variance and Model Complexity7.3 The Bias-Variance Decomposition7.3.1 Example: Bias-Variance Tradeoff7.4 Optimism of the Training Error Rate7.5 Estimates of In-Sample Prediction Error7.6 The Effective Number of Parameters7.7 The Bayesian Approach and BIC7.8 Minimum Description Length7.9 Vapnik Chernovenkis Dimension7.9.1 Example (Continued)7.10 Cross-Validation7.11 Bootstrap Methods7.11.1 Example (Continued)Bibliographic NotesExercises8 Model Inference and Averaging8.1 Introduction8.2 The Bootstrap and Maximum Likelihood Methods8.2.1 A Smoothing Example8.2.2 Maximum Likelihood Inference8.2.3 Bootstrap versus Maximum Likelihood8.3 Bayesian Methods8.4 Relationship Between the Bootstrap and Bayesian Inference8.5 The EM Algorithm8.5.1 Two-Component Mixture Model8.5.2 The EM Algorithm in General8.5.3 EM as a Maximization-Maximization Procedure 8.6 MCMC for Sampling from the Posterior8.7 Bagging8.7.1 Example: Trees with Simulated Data8.8 Model Averaging and Stacking8.9 Stochastic Search: BumpingBibliographic NotesExercises9 Additive Models, Trees, and Related Methods9.1 Generalized Additive Models9.1.1 Fitting Additive Models9.1.2 Example: Additive Logistic Regression9.1.3 Summary9.2 Tree Based Methods 10 Boosting and Additive Trees11 Neural Networks12 Support Vector Machines and Flexible Discriminants13 Prototype Methods and Nearest-Neighbors14 Unsupervised LearningReferencesAuthor IndexIndex

插图：

《统计学习基础》由世界图书出版公司出版。

无

统计学习基础 PDF格式下载

从网站给出的信息来看，这是第一版，但我有Emule下载的是第二版，第二版2008年出版，如果是这样的话，世图太差了，有新版本却影印老版本。

买回来才发现，比较崩溃……

虽然这是第一版的书，比第二版的少四章，但我没找到第二版有卖的，所以即使是第一版还是很不错的，不要以为出了第二版就完全否定第一版，买这本书的人谁把里面的内容都掌握了才是最重要的

这本书是我学习机器学习的入门书。书中基本上包含了大部分的机器学习算法。内容翔实，数学证明充分。刚刚开始的时候读起来有点难，读久了就觉得这本书特别好，数学证明充分，基本功可以练的很扎实。有的时候，自己能安静下来，认真把一些书中没有讲到的证明慢慢写出来，挺有收获。

书纸张质量很差，而且每装订好。很容易就散开，脱页。在亚马逊买这么多东西，就这次最不满意。以后买书要掂量下，还是去当当买比较好。

书脊的装订对不起印刷的质量，不敢用力压书，生怕从中间断开。书的内容相当不错，和PRML比起来，没有像后者一样把所有的内容都统一进概率的框架里面去解释。

经典必读,无须多说!

好评太多，忍不住就买了，挺不错的，质量很好

居然是彩印的，书的内容，手感都很好！

翻译的作品有时候能误导读者，所以对照着读效果更好！

印刷质量不错，不过书太厚了，好像我的就要从中间断开了。其他都很不错，是一本经典的书。要是想看那些修正，从网站上也有。

书质量很好。内容经典，值得推荐

虽然是第一版，彩色印刷还是很喜欢。

质量很不错，插图颜色很鲜明，书的气息让我想起了当年读英文原版的哈利波特

印刷精美，不愧是经典作品！

送书的人态度比较差。。。特别捉鸡

一个很好的书

统计学习基础

好东东，送货快，价格公道。

可以可以可以啊啊可以可以可

中间开线了

　　中文翻译版大概是用google翻译翻的，然后排版一下，就出版了。所以中文翻译版中，每个单词翻译是对的，但一句话连起来却怎么也看不懂。最佳阅读方式是，看英文版，个别单词不认识的话，再看中文版对应的那个词。但如果英文版整个句子都不懂的话，那只有去借助baidu/google，并运用联想、推理能力来自己理解了。

　　个人觉得“机器学习 -- 从入门到精通”可以作为这本书的副标题。
　　
　　机器学习、数据挖掘或者模式识别领域有几本非常流行的教材，比如Duda的模式分类，Bishop的PRML。Duda的书第一版是模式识别的奠基之作，现在大家谈论得是第二版，因为内容相对简单，非常流行，但对近20年取得统治地位的SVM、Boosting基本没提，有挂一漏万之憾。PRML侧重概率模型，体系详备，是Bayesian方法的扛鼎之作。和PRML相比，这本Elements of Statistical Learning对当前最为流行的方法有比较全面深入的介绍，对工程人员参考价值也许要更大一点。另一方面，它不仅总结了已经成熟了的一些技术，而且对尚在发展中的一些议题也有简明扼要的论述。让读者充分体会到机器学习是一个仍然非常活跃的研究领域，应该会让学术研究人员也有常读常新的感受。
　　
　　这本书的作者是Boosting方法最活跃的几个研究人员，发明的Gradient Boosting提出了理解Boosting方法的新角度，极大扩展了Boosting方法的应用范围。书中Boosting部分是被相关学术论文引用最频繁的部分。个人觉得经常研读一下作者和其他Boosting流派打嘴仗的文章是学习机器学习很好的一个途径，因为只有这样尚未成熟(而又影响广泛)的领域中，你才能更具体地体会到一个学科是怎样逐渐发展成熟的，那些贡献卓著的研究人员是如何天才地发现问题解决问题的，又是如何因偏执而终究会被证明有一方至少是部分地无知的。这种体会是很难在那些发展成熟了的分支中找到的。Regularization方法是作者贡献丰富的另一个领域，也是这本书另一个最具趣味的部分。
　　
　　这本书第一版在2000年出版，现在评论的第二版是09年出版的，包含了很多值得玩味的新内容。比如从Ensemble方法的角度来解释MCMC方法的优异性能，就是我以前没有注意到的。当然，也许只是因为我的知识范围还不够宽。
　　
　　
　　

　　这本统计学习的书由斯坦福几个响当当的大牛所写，覆盖面很广且阐述的比较透彻，一些最新的（2008/2009）研究成果也收录其中,能够给读者对统计学习领域一个全面、清晰的认识。统计和生统行当的必备道具，如果你做这些行当，千万别跟同行说不知道这本书。。。

而且还缺了一些内容！！！！！！

同感，可以当字典查单词，很多地方倒不如英文的来的明白啊

说说我的感受
全书大量使用Regularization Operator和Sampling，却没有high level的理论分析，实在意犹未尽
另外对Non-Flatten数据的处理太少

怎么我记得对Regularization从最大后验和SVD两个角度解释了？是在别的书里看到的？
Non-Flatten您是说manifold吗？这个的确不是本书重点。

　　Learning with Kernels谈了Regularization Operator和RKHS的关系，范剑青等人讨论了SCAD等其他Norm
　　Regularization的解释是加入一些常识性的问题理解吧，比如通常是光滑的、稀疏的之类，感觉和概率那套先验后验还是不大一样
　　我也没有细看，随便扯两句啊 呵呵
　　
　　Non-Flatten就是Structural、Relational、Hierarchical之类的

是啊，模型通常是光滑稀疏的，这是一个先验知识啊，比如依据上面这个先验知识令参数高斯分布，那么后验就得到L2的Regularization。Learning with Kernels没认真看过。

一看Regularization，总是让我想起Tomaso Poggio

"比如从Ensemble方法的角度来解释MCMC方法的优异性能"
这是在哪个章节？