Statistical Methods for Spatial Data Analysis
Statistical Methods for Spatial Data Analysis
复制标题
DOI：
10.1198/jasa.2006.s66
复制
发表时间：
2006-03
影响因子：
3.7
通讯作者：
Jun Zhu
中科院分区：
文献类型：
--
作者：
Jun Zhu
这本书组织得很好。四个主要章节按逻辑顺序排序，从基础理论、计算算法到高级应用。一些带星号的部分是高级部分，可以跳过。本书内容紧凑，很容易让人产生成就感和进步感。特别是，我很高兴有两个主题没有被涵盖。其中之一涉及计算机科学文献中的重新排序技术的细节。作者“建议将构造和实现稀疏矩阵分解算法的问题留给数值和计算机科学专家”（第 52 页）。另一个涉及MCMC的详细内容，可以在很多现有书籍中找到。本书的紧凑性与那些试图过于完整而最终厚得令人生畏的书不同。这本书有一个专门的网站（http://www.math.ntnu.no/̃hrue/GMRFbook/），其中包含数据集、软件和其他有用的材料。本书的配套软件是一个 C 库 GMRFlib，属于公共领域。该库提供了实现书中描述的算法的函数：GMRF的无条件模拟、GMRF的各种类型的条件模拟、相应对数密度的评估以及MCMC算法中块更新的生成。按照文档中的说明，我毫无困难地安装了该库并成功运行了示例代码。如果这些设施可以在高级计算环境（例如 R）中使用，那就太好了，但是对具有 GMRF 组件的典型 MCMC 应用程序的速度的担忧可能会促使人们重新在较低级别的编译代码中实现。我认为值得投入时间来学习如何使用该库，在线文档使这变得更加容易。总而言之，我可以推荐这本书作为一本非常好的研究生水平教科书。我也相信读者会从 GMRFlib 软件中学到很多东西。
The book is well organized. The four major chapters are sorted in a logical order, from basic theory, computational algorithms, to advanced applications. Some starred sections are advanced and may be skipped. The book is compact, making it easy to extract a feeling of accomplishment and progress. In particular, I appreciated that two topics are not covered. One involves details of the reordering techniques from the computer science literature. The authors “recommend leaving the issue of constructing and implementing algorithms for factorizing sparse matrices to the numerical and computer science experts” (p. 52). The other involves details on MCMC, which can be found in many existing books. The compactness distinguishes the book from those which try to be too complete and end up being intimidatingly thick. The book has a dedicated website (http://www.math.ntnu.no/ ̃hrue/GMRFbook/ ) which contains datasets, software, and other useful materials. The software companion of the book, a C library GMRFlib, is in the public-domain. This library provides functions implementing the algorithms described in the book: unconditional simulation of a GMRF, various types of conditional simulation from a GMRF, evaluation of the corresponding log-density, and generation of block updates in MCMC-algorithms. Following the instructions in the document, I installed the library without difficulty and ran the example code successfully. It would be nice if these facilities were available in a high-level computing environment, such as R, but concerns about speed in typical MCMC applications with GMRF components would likely drive one back to implementing in lower-level compiled codes. I think it is worth investing the time to learn how to use the library, which is made easier by the online document. In summary, I can recommend this book as a very good graduate level textbook. I also believe that readers will learn much from the GMRFlib software.