博文
【读书笔记】Support vector machines : theory and application
|
本书是将同支持向量机相关的各种扩充工作进行了编著。开篇的介绍写得清晰、实用，而后面的章节则各自为题，良莠不齐。不适合做为入门书籍，或者教学课本，而仅仅当做对支持向量机有一定了解的基础上的思维拓展。例如前面几章有定义新的目标函数或优化算法，大多有同其他的统计学习方法相结合，提供更“合理”的分类模型。而全书后面的几章都是偏应用的课题，感觉比较混乱。虽然这样的安排切合了本书的书名——理论和应用，但是由于理论和应用的章节之间并没有联系，所以给读者帮助有限。
书中有些段落还不错，摘录如下：
Page 44:
A training and design of a support vector machine is an iterative algorithm and it involves the following steps:
(a) define your problem as the classification or as the regression one,
(b) preprocess your input data: select the most relevant features, scale the data between [-1,1], or to the ones having zero mean and variances equal to one, check for possible outliers (strange data points),
(c) select the kernel function that determines the hypothesis space of the decision and regression function in the classification and regression problems respectively,
(d) select the "shape", i.e., "smoothing" parameter of the kernel function (for example, polynomial degree for polynomials and variances of the Gaussian RBF kernels respectively),
(e) choose the penalty factor C and, in the regression, select the desired accuracy by defining the insensitivity zone epsilon too,
(f) solve the QP problem in l and 2l variables in the case of classification and regression problems respectively,
(g) validate the model obtained on some previously, during the training, unseen test data, and if not pleased iterate between steps (d) (or, eventually c) and (g).
Page 115:
Global learning summarizes the data and provides the practitioners with knowledge on the structure of data, since with the precise modeling phenomena, the observations can be accurately regenerated and therefore thoroughly studied or analyzed. However, this also presents difficulties in how to choose a valid model to describe all the information. In comparison, local learning directly employs part of information critical for the specifically oriented tasks and does not assume a model for the data. Although demonstrated to be superior to global learning in various machine learning tasks, it misses some critical global information.
图书信息：
https://blog.sciencenet.cn/blog-286702-600892.html
上一篇：【读书笔记】Statistical bioinformatics with R
下一篇：【读书笔记】Probabilistic methods for bioinformatics with
书中有些段落还不错，摘录如下：
Page 44:
A training and design of a support vector machine is an iterative algorithm and it involves the following steps:
(a) define your problem as the classification or as the regression one,
(b) preprocess your input data: select the most relevant features, scale the data between [-1,1], or to the ones having zero mean and variances equal to one, check for possible outliers (strange data points),
(c) select the kernel function that determines the hypothesis space of the decision and regression function in the classification and regression problems respectively,
(d) select the "shape", i.e., "smoothing" parameter of the kernel function (for example, polynomial degree for polynomials and variances of the Gaussian RBF kernels respectively),
(e) choose the penalty factor C and, in the regression, select the desired accuracy by defining the insensitivity zone epsilon too,
(f) solve the QP problem in l and 2l variables in the case of classification and regression problems respectively,
(g) validate the model obtained on some previously, during the training, unseen test data, and if not pleased iterate between steps (d) (or, eventually c) and (g).
Page 115:
Global learning summarizes the data and provides the practitioners with knowledge on the structure of data, since with the precise modeling phenomena, the observations can be accurately regenerated and therefore thoroughly studied or analyzed. However, this also presents difficulties in how to choose a valid model to describe all the information. In comparison, local learning directly employs part of information critical for the specifically oriented tasks and does not assume a model for the data. Although demonstrated to be superior to global learning in various machine learning tasks, it misses some critical global information.
图书信息：
| Title | Support vector machines : theory and applications / Lipo Wang (ed.). |
| Publisher | Berlin : Springer, 2005. |
https://blog.sciencenet.cn/blog-286702-600892.html
上一篇：【读书笔记】Statistical bioinformatics with R
下一篇：【读书笔记】Probabilistic methods for bioinformatics with
