Bootstrap是通过随机模拟扩充小样本的方法

非统计专业人士使用某统计方法最先应该知道的是，哪些情况下这种方法不能使用、以及这方法有什么特别的局限性。然后就可以放心大胆拿来就用了。

bootstrap得到的参数估计之类的结果，都是随机变量（不具有严格的可重复性）。这在我看来是最大的缺点。

数据或样本太少了要做统计参数的估计怎么办？对样本服从的分布情况缺少先验的认识怎么办？

Bootstrap方法，基于同一个数据或样本、独立自主自力更生、随机再抽样地反复、大量扩充样本。

可以实现这种方法的软件工具很多。matlab中主要是bootci 和bootstrp 两个函数。

此bootstrap非彼bootstrap , 这里说的只是一种统计学中的基于monte carlo模拟的方法，跟软件和网络开发无关。

除了CRC出版的Computational Statistics Handbook with MATLAB® （最新2015第三版，网络上容易找到Second Edition，并可下载工具箱代码），里面有Bootstrap及Jackknife方法的详细介绍、实施的代码等细节。

还有学者不到20年前开源的一个工具箱介绍如下（也在上面的书中有提到）：

http://www.spg.tu-darmstadt.de/res/dl/bootstraptoolboxformatlab/bootstrapmatlabtoolbox.en.jsp

BOOTSTRAP MATLAB TOOLBOX

下载链接

Abdelhak M. Zoubir

D. Robert Iskander

CSP Group,

Curtin University of Technology

Introduction

In many signal and information processing applications one is interested in forming estimates of a certain number of unknown parameters of a random process, using a set of sample values. Further, one is interested in finding the sampling distribution of the estimators, so that the respective means, variances, and cumulants can be calculated, or in making some kind of probability statements with respect to unknown true values of the parameters. For example one could be interested in assigning two limits to a certain parameter, and in asserting that, with some specified probability, the true value of the parameter will be situated between these limits, which constitute the confidence interval.

The bootstrap is a powerful technique for assessing the accuracy of a parameter estimator in situations where conventional techniques are not valid. The bootstrap does with a computer what the experimenter would do in practice if it were possible: he or she would repeat the experiment. With the bootstrap, the observations are randomly reassigned, and estimates recomputed. These assignments and recomputations are done thousands of times and treated as repeated experiments.

The Bootstrap Toolbox

The Bootstrap Toolbox is a set of Matlab functions consisting of procedures for resampling, hypothesis testing, and confidence interval estimation. Each function of the Toolbox has a help entry. See the Bootstrap Reference Manual for details.

Examples from an article published in the IEEE Signal Processing Magazine on the application of the bootstrap to signal processing [1] are also provided. More examples of the application of bootstrap techniques can be found in [2,3,4,5,6].

The Toolbox runs under Matlab 5.0 (or higher) and requires the Statistical Matlab Toolbox, Version 2.1.0 (or higher). Some functions may run under Matlab v4.2c.

The Bootstrap Toolbox has been developed by Abdelhak M. Zoubir and D. Robert Iskander. Hwa-Tung Ong is greatly acknowledged for his help and comments during the development of the Toolbox.

Registration

All users of the Bootstrap Toolbox are encouraged to register themselves so they can be notified about upgrades and enhancements.

To register, e-mail: zoubir@ieee.org with subject line “Bootstrap registration”.建议下载之后用此邮件标题注册。发邮件到左边邮箱即可。

Download the Bootstrap Toolbox

可以自由使用的matlab代码，81KB大小，可以从这里下载。

The Bootstrap Toolbox is available FOR FREE at: bootstrap.zip (81kb)

The package contains a set of M-files, and a postscript version of the Bootstrap Reference Manual.

Warranty and Bugs

There is NO WARRANTY attached to this software. If you find that it does not work correctly, please compile a description of the Matlab code that generates the error. E-mail the description tozoubir@ieee.org. An effort will be made to correct the problem for a future release.

References

• A. M. Zoubir and B.Boashash, “The Bootstrap and Its Application in Signal Processing ”, IEEE Signal Processing Magazine, 15(1):56-76, 1998.

• A. M. Zoubir and D. R. Iskander, “Bootstrapping Bispectra: An Application to Testing for Departure from Gaussianity of Stationary Signals”, IEEE Transactions on Signal Processing, 47(3):880-884, 1999.

• A. M. Zoubir and J. F. Bohme, “Bootstrap Multiple Tests Applied to Sensor Location”, IEEE Transactions on Signal Processing, 43(6):1386-1396, 1995.

• A. M. Zoubir, “Bootstrap Multiple Tests: An Application to Optimum Sensor Location for Knock Detection ”, Applied Signal Processing, 1:120-130, 1994.

• A. M. Zoubir, “Bootstrap: Theory and Applications ”, In F. T. Luk, editor,Advanced Signal Processing Algorithms, Architectures and Implementations, volume 2027, pages 216-235, San Diego, USA, July 1993. Proceedings of SPIE.

MATLAB is a registered trademark of The MathWorks, Inc.