Algorithms for Graphical Models (AGM)

[Announcements] [Timetable] [Lecture schedule with links to corresponding online material] [Practicals] [Textbooks] [gPy and the algorithm library] [Assessment] [Acknowledgements] [Other useful links]


Announcements


Timetable

Type Day Time Place Duration
Lecture Tuesday 0915-1015 LMB/031 Weeks: 02-10
Lecture Tuesday 1515-1615 RCH/Lakehouse Weeks: 02-10
Practical Friday 0915-1115 CSE/069 Weeks: 03-10
Surgery Tuesday 1015-1115 Room 326, The Hub Weeks: 02-10

Lecture schedule

No. Title Materials Day Cussens Cowell et al
01 Introduction HERE 02/Tue/0915
02 Python: basics HERE 02/Tue/1515
03 Python: object-orientation HERE 03/Tue/0915
04 Data and probabilities HERE 03/Tue/1515 3.1 - 3.2 2.5 - 2.8
05 Factors HERE 04/Tue/0915 4.1 - 4.5 6.2
06 Conditional independence in factored distributions HERE 04/Tue/1515 5.1 - 5.3 4.1, 5.1, 5.2
07 Variable elimination HERE 05/Tue/0915 4.6 - 4.10, 5.4 4.8 - 4.10
08 Bayesian nets HERE 05/Tue/1515 7.1 5.3
09 Decomposable models and join forests HERE 06/Tue/0915 6.2 4.2 - 4.4
10 Probability propagation in join forests HERE 06/Tue/1515 6.1, 6.3 6.1 - 6.3
11 07/Tue/0915
12 Rejection and importance sampling HERE 07/Tue/1515 8.1 - 8.5
13 Gibbs sampling HERE 08/Tue/0915 8.6 Appendix B
14 Bayesian parameter estimation HERE 08/Tue/1515 9.1 - 9.4
16 Dynamic graphical models HERE 09/Tue/0915
17 Structure learning HERE 10/Tue/0915 Chapter 11
18 Spare/revision 10/Tue/1515

Practicals

No. Title Day
01 Python programming 3/Fri
02 Factored distributions, graphs and hypergraphs 4/Fri
03 Knowledge engineering with Bayesian nets 5/Fri
04 Graphs and hypergraphs 6/Fri
05 Join trees and sampling 7/Fri
06 Gibbs sampling with BUGS 8/Fri
07 Parameter estimation 9/Fri
08 Machine learning 9/Fri

Textbooks

An accompanying textbook Algorithms for Graphical Models is available. Feel free to send me comments on it.

A useful (complete!) textbook is Probabilistic Networks and Expert Systems by Cowell, Dawid, Lauritzen and Spiegelhalter. There are 5 copies in the library available under a variety of loan schemes.


gPy and the algorithm library

gPy is a collection of Python modules (a Python package) which support the teaching of this module. To be able to use them in your Python programs it suffices to set the environmental variable PYTHONPATH to /usr/course/agm/gPy. If you use the (default) bash shell you can do this by adding the following line to your .profile in your home directory.

export PYTHONPATH=/usr/course/agm/gPy

The environmental variable PYTHONPATH will then be set to the right value each time you log in. Typing source .profile will set it immediately.

If you use tcsh add

setenv PYTHONPATH /usr/course/agm/gPy

to your .tcshrc file.

There is API documentation for gPy.

In the table below, the links for each algorithm take you to its top-level description. Clicking on "source code" in the top right-hand corner of this description takes you to the source which may be annotated. Clicking on highlighted names in the source will take you to the top-level description of the object in question (usually a method).

Algorithm Author(s) Annotated Lecture Comments
Factor marginalisation TODO 05 Low-level, gPy-specific
Applying a binary operation (e.g. multiplication) between two factors TODO 05 Low-level, gPy-specific
Finding the cliques of a graph Bron & Kerbosch Partially
Gibbs sampling 13
Importance sampling 12
Moralisation 08
Triangulation with a given elimination ordering 09
Iterative proportional fitting 15
Join forest model calibration Lauritzen & Spiegelhalter 10
Join forest construction from an arbitrary hypergraph and an elimination ordering TODO 10
Join forest construction from a decomposable hypergraph using vertex elimination Graham; Beeri et al YES 09
Finding a perfect sequence of cliques in a join forest TODO 10
Rejection sampling 12
Maximum cardinality search for graphs Tarjan & Yannakakis YES 09
Restricted maximum cardinality search on hypergraphs Tarjan & Yannakakis NO 09
Variable elimination 07
Constructing the essential graph of a digraph Chickering TODO

Assessment

Assessments from previous years


Acknowledgements


Other useful links

Internal

Graphical models

Python


Last modified: Tue Dec 6 16:58:55 GMT 2011

Valid XHTML 1.1!