Mattias Villani
2013-11-20
Division of Statistics
Dept. of Computer and Information Science
Linköping University
mattias.villani@liu.se
Bayesian Learning, 6 hp
Guidelines for the project
This is a document with some rough guidelines for the project in the Bayesian learning
course. Note however that the project is very exible and I don't want to give too rigid
guidelines for it.
• Deadline:
December 22. It is ok to submit earlier. :-)
• Individual.
The project is to be done individually.
You can of course have the
usual discussions with classmates about the Bayesian learning in general (I assume
this is constantly on your mind!), but don't ask for specic solutions to problems
that you encounter in your project. It is YOUR project, YOU get the grade, YOU
do the work.
• The product:
PDF format
no oral
concise. Explain
You are expected to send me a written report in
(Word or other word processing software are not acceptable). There will be
presentation.
The report should be
clearly written
and be
all the things you do so that I can see that you actually understood what you did
in the report.
Pushing a button in some package and dumping a lot of output
in the report without proper explaination and interpretation is not good enough.
Make sure that you use the proper Bayesian terminology that we have used during
the course. It is impossible to say anything sensible about the extent (how many
pages?) of the report. All I can say is that the project is 4 credit points and it is
the only examination for the course. Attach computer code in an appendix of the
project report.
• Software.
You are free to use any available software, but using pre-packaged soft-
ware rather than your own written program code increases the demands on the statistical analysis and modelling, and makes it even more important that you clearly
interpret the results to show that you have understood the theory and concepts
underlying the analysis performed by the software package.
• The content of the project.
You are free to work on any problem you like, but I do
expect you to do a Bayesian analysis of the problem. The typical project involves
the following steps:
A well-dened
problem denition.
The problem denition can change when
you start analyzing the data, but having an idea to start with helps. I would
suggest that you pick a problem that YOU care about. Make it fun for you.
There is no need to make any kind of novel scientic contribution in this course
(but I don't mind if you do!).
1
A
dataset.
It is ok to look at a theoretical problem without any data, but
most students prefer to analyze some data. It is certainly ok to use several data
sets, and even to divide the project report in more or less unrelated parts. But
I think that it is easier to do a good project if you focus on a single dataset.
Models.
When you choose a problem denition and a dataset, make sure that
it gives you the oppurtunity to analyze interesting models. You can try several
models of dierent parts of the data set (see for example in my slides how I rst
analyzed the Spam data set using a simple Bernoulli model with Beta prior,
and later on introduced covariates and used logistic regression).
Analysis.
Once you have your dataset and some estimated models, move
on to analyze the results of your analysis.
Maybe you can illustrate how to
do predictions with the model (perhaps computing the predictive distribution
by simulation)? How to evaluate the model t? Analyze which covariates to
include in the model, if you are in a regression-type of situation?
• Project proposal:
30.
Submit a project proposal to me no later than
November,
The project proposal is a very short written document (maximum a page or
so) where you explain what you intended to do in the project. Use the four bullet
points above (problem denition, dataset, models and analysis) to structure your
project proposal. Keep it brief, just tell me enough so that I can judge if the project
is worth pursuing.
• Oral examination.
For master students (which are graded on the A-F scale) I may
need to complement the project report with oral examinations for those students for
which I am uncertain about the grade. Oral examinations will be planned bilaterally
between me and the selected students, but will take place sometimes in the latter
part of January.
Good luck!
be ambitious.
You have plenty of time later in life to be a slacker.
2