mcmc代码蒙特卡洛马尔科夫链

MCMC Methods for MLP and GP and Stuff (for Matlab 6.*) Maintainer: Aki Vehtari <Aki.Vehtari@hut.fi> This software is distributed under the GNU General Public Licence (version 2 or later); please refer to the file Licence.txt, included with the software, for details. Table of contents: 1. INTRODUCTION 2. DOCUMENTATION AND DEMOS 3. INSTALLING THE TOOLBOX ------------------------------------------ 1. INTRODUCTION Basic design of this toolbox is based on Netlab <http://www.ncrg.aston.ac.uk/netlab/>. However, this toolbox is not compatible with Netlab, because the option handling has been changed to use structures similar to current default in Mathworks' toolboxes. Furthermore, the code in this toolbox has been streamlined and optimized for faster computation, and it has been extended to include some of the features present in FBM <http://www.cs.toronto.edu/~radford/fbm.software.html> and some other features. Some of the most computationally critical parts have been coded in C. For easier introduction to MLP's and GP's, Netlab is better suited, especially because of the accompanying text book (Nabney I.T. (2001) Netlab: Algorithms for Pattern recognition). Purpose of this toolbox was to port some of the features in the FBM to Matlab environment for easier development for Matlab users. Most of the code has been written by Aki Vehtari in the Laboratory of Computational Engineering <http://www.lce.hut.fi> at Helsinki University of Technology (HUT) <http://www.hut.fi>. Currently there is also code written by (in alphabetical order) Toni Auranen, Christopher M Bishop, James P. LeSage, Ian T Nabney, Radford Neal, Carl Edward Rasmussen, and Simo S�rkk�. During the summer of 2004 the code was fully documented and pulled together as a toolbox by Jarno Vanhatalo, an undergraduate student at HUT. Speciall thanks are directed to Prof. Jouko Lampinen who helped in converting the C sources into mex-files for Windows. 2. DOCUMENTATION AND DEMOS: In the toolbox there are two demonstration programs, that solve a two dimensional regression problem with MLP network and Gaussian process. These are demo_2input and demo_2ingp. The demos are discussed in more detail in the home page of the toolbox <http://www.lce.hut.fi/research/compinf/mcmcstuff/demos> and in documentation of the software that is available from the home page of the toolbox <http://www.lce.hut.fi/research/compinf/mcmcstuff/>. Features treated in the documentation are: - Baysian learning for MLP in a regression problem - a Gaussian Hierarchical prior structure with Automatic Relevance Determination (ARD) - noise model with normal distribution - Markov Chain Monte Carlo method - hybrid Monte Carlo and Gibbs sampling - a Gaussian process in a regression problem - Bayesian learning for MLP in classification problems with two or many classes - a Gaussian process in classification problem with two classes - noise model with Students t-distribution - input variable selection for MLP and GP (various methods) Features not treated in the documentation are: - covariate dependent groupped noise model 3. INSTALLING THE TOOLBOX -All files needed to run the toolbox in linux, Alfa and Windows (*.m, *.mexaxp, *.mexglx, *.dll) are located in the same directory with this Readme.txt (mcmcstuff/). The toolbox should work in the above architechtures after the mcmcstuff/ is added to Matlab path. -In the directory mcmcstuff/ there are also two subdirectories linuxCsource/ and winCsource/ containing the functions written in C for Linux, Alfa (linuxCsource/) and Windows (winCsource/). The reason for two different folders is that the 'math.h' of C library is different in Linux and Windows and that the function names in BLAS libraries are different in Windows than in Linux and Alfa. For example in Linux 'dpotrf_' is in Windows 'dpotrf'. -The toolbox can be tested by running demonstration program demo_2ingp (NB! the demo takes quit a long time, approximately 30 minutes on a 2400MHz Intel Pentium workstation, it is sufficient enough to run the demo for few minutes to be sure that it works). If there is no right mex-files for the architechture of your computer, you get an error message "error, No mex-file for this archtitecture. See Matlab help and convert.m in ./linuxCsource or ./winCsource for help." -There are two different compiler m-files, compile.m in mcmcstuff/linuxCsource/ and compile_windows.m in mcmcstuff/winCsource/. The *.c files can be converted to mex-files by running these files in Matlab in the same directory where the it is located. If the compilers do not work directly the relations between C-functions can still be seen from them.