Instructor: Kevin Patrick Murphy http://www.cs.ubc.ca/~murphyk/

TA: Frank Hutter http://www.cs.ubc.ca/~hutter

Course Webpage: http://www.cs.ubc.ca/~murphyk/Teaching/CS540_Fall05/index.html

Final Project:

Performance comparison of approximate inference techniques used by WinBugs & Vibes.

WinBugs Linear Regression (1-D & 2-D):

1. 1-D model, data & initialization (.zip) < script file >

2. 2-D uncorrelated weights - model, data & initialization < script file >

3. 2-D correlated weights - model, data & initialization (.zip)< script file >

Instructions for running codes:

Download the .zip files, extract them in the WinBugs directory (commonly C:\Program Files\WinBUGS14). Download the corresponding script files & save them in the same directory. Type in the command:

WinBUGS14.exe /PAR <script file name>

For example to run the demo for 2-D correlated weights, type

WinBUGS14.exe /PAR script_2d_cor.txt

in the WinBUGS directory. See WinBugs webpage for instructions & details on its usage at: http://www.mrc-bsu.cam.ac.uk/bugs/

Vibes Linear Regression (2-D):

1. Matlab Codes (.zip)

Instructions for running the codes:

Since Vibes is free, I have bundled it with the zip file. Extract it at a convenient place & run the functions. This function can only be run in Matlab 7.0, Release 14 & does not work with Matlab 6 & lower versions (trust me, I spent several days trying to make it work in Matlab 6). If you get an error like:

> In vlinregr at 65
??? The function, script, or class Vibes cannot be indexed using {} or . indexing.

It means Vibes JAR file has not registered itself with Matlab. This is a known issue. I suggest you try multiple times with intervals in between each attempt till the JAR gets initialized (and yes, eventually it should).

For any other query check Vibes webpage at:  http://vibes.sourceforge.net/index.shtml

Project Presentation Slides

1. PDF <<6-up>> <<1-up>>

2. Powerpoint PPT

Page Last Updated: