An Introduction to Sequential Monte Carlo Methods
An Introduction to Sequential Monte Carlo Methods
复制标题
DOI：
10.1046/j.1467-9884.2003.t01-6-00383_8.x
复制
发表时间：
2003-12
期刊：
影响因子：
--
通讯作者：
Freda Kemp
中科院分区：
文献类型：
--
作者：
Freda Kemp
这类似于一个宣言，真的需要与Christakos（2000）一起阅读才能获得最佳效果。作者试图对比他们的方法，贝叶斯最大熵的基础上，这需要一些基本的物理模型的理解，与“统计”方法，如Diggleet等人开发的。（1998年）。例如，克里金法是他们所提倡的方法的一个特例，适用于某些狭窄的条件。这在实践中似乎是一个相当突然的区别，特别是考虑到作者后来的断言（第192页），其中承认科学家可能会找到一些最适合研究某些部分的方法。然而，这种对比有助于区分书中所包含的材料。本书的部分哲学性质在序言中得到承认，但作者陈述了他们专注于“关键实用性”的意图，他们将其定义为使用时间地理信息系统来更好地科学理解正在调查的现象，提供准确的预测并做出有用的决定。本书的第7章介绍了BME-Lib，这是一套MATLAB软件，也包含在本书的光盘上。同样，需要熟悉Christakos（2000）才能理解BMELib中包含的方法。这显然是一本专门的书。它没有声称其目标读者，但似乎暗示它是针对时间地理信息系统专家的，并表示希望他们能够在更基本的科学层面上与他们所涉及的各种科学专家进行交流。所提倡的方法的一部分是最大限度地利用所研究现象的物理模型。所提供的软件需要MATLAB 5.0或更新版本（这显然排除了Octave等应用程序），并且应该在任何可以运行MATLAB的环境下运行。可修改的示例脚本包含在软件中，使其易于使用，因为它可以给出一章书面摘要。引用文献163篇，其中32篇是作者自己的作品。索引是适度的，印刷在不到三页，这可能反映了大部分的书的哲学性质。作者认为，使用这种方法的情况下，而不是给一个非常清楚的阐述，无论是理论或详细的案例研究使用的方法。然而，它提出了一个发人深省的论点，关于科学方法的性质，这是感兴趣的本身（它是特别严厉的简单依赖奈曼皮尔逊假设检验和损害时，不同群体声称对比结果的公众对科学的看法）。然而，这可能是不够的兴趣，在这样一个专门的书值得购买这一点。所取得的是一个强制提出的拥护双方的作者的整体方法和他们的研究，可能感兴趣的信息科学家与时空现象的工作。MATLAB软件的提供为时态地理信息系统提供了贝叶斯最大熵方法的实现，对于那些对其应用感兴趣的人来说是有价值的。
This resembles a manifesto and really needs to be read in conjunction with Christakos (2000) for best effect. The authors try to contrast their methodology, based on Bayesian maximum entropy and which requires some understanding of the underlying physical models, with ‘statistical’methods such as those developed by Diggleet al.(1998). For example, kriging is presented as a special case of the approach that they advocate, which is applicable under certain narrow conditions. This seems a rather abrupt distinction in practice, especially given the authors’ assertion later (page 192) in which it is acknowledged that scientists may find some approaches most suitable for certain parts of an investigation. However, this contrast serves to differentiate the material that is contained in the book. The philosophical nature of parts of the book is acknowledged in the preface, but the authors state their intention of focusing on ‘critical practicality’which they define as using temporal geographical information systems to gain a better scientific understanding of the phenomena under investigation, to provide accurate predictions and to make useful decisions. Chapter 7 of the book introduces BME-Lib, a set of MATLAB software also included on a compact disc with the book. Again, familiarity with Christakos (2000) is required to understand the methodology that is contained within BMELib. Clearly this is a specialized book. It makes no claims about an intended audience but seems to imply that it is aimed at temporal geographical information systems specialists and expresses the hope that they may be able to communicate at a more fundamental scientific level with the various scientific specialists with whom they are involved. Part of the approach that is advocated is to maximize the use of physical models of the phenomena under study. The software that is provided requires MATLAB 5.0 or newer (which obviously rules out applications such as Octave) and should run under any environment which can run MATLAB. Modifiable example scripts are included with the software which make it as easy to use as it could be given a one-chapter written summary. 163 references are cited, 32 of which are to the authors’ own work. Indexing is modest, printed on fewer than three pages, which may reflect the philosophical nature of much of the book. The authors argue the case for using this type of methodology rather more than for giving a very clear exposition of either the theory or detailed case-studies using the methodology. Nevertheless it presents a thoughtprovoking argument about the nature of the scientific method which is of interest in its own right (it is particularly scathing about simple reliance on Neyman–Pearson hypothesis testing and the damage that is done to public perceptions of science when different groups claim contrasting results). However, this is probably not of enough interest in such a specialized book to merit purchase for this alone. What is achieved is a forcibly presented espousal of both the authors’ overall approach and their research that may be of interest to information scientists working with spatiotemporal phenomena. The provision of MATLAB software provides a realization of the Bayesian maximum entropy approach to temporal geographical information systems and is valuable for those who are interested in its application.