Messages
The syllabus has now been updated, see /studier/emner/matnat/math/STK4021/h22/pensum/
Some of you might have got some results that looks strange for the models in problem 3 (all effects seems to be zero).
However, this seems mainly to be due to that the effects become small (but still significant). By using the option digits=5 either in the print or summary commands on the object you get, you will be able to see the results. Further, if your object is fit.Bayes, you may get a better plot of the random effects by using the command
plot(fit5.Bayes,pars="b")
Geir
The last part about comparing waic (or loo) is a bit tricky and is ok if you skip this point. For those interested, note that the pointwise object gives elppd. In order to get lppd you need to add p_waic (see eq (7.13) in the book). Then you can adjust the individual lppd values according to the transformation before you subtract p_waic again in order to get the elppd value on the right scale.
There has been some questions about 1c which assumes an Inv-Gamma prior for sigma^2 while in the practical part the default prior is to be used.
The main reason for this is that it becomes a bit technical to use the Inv-Gamma within the stan_glm routine, therefore just stick with the default choice.
In order to cover the topics needed for the compulsory exercise as soon as possible, we have two hours lecture today and leave the exercises for Wednesday.
The compulsory exercise is available here. This exercise covers linear regression settings which we have not discussed much so far in the course (but which we will discuss shortly). Almost all commands needed are however provided and I advice you to start as soon as possible to try out these and get it to work.
I have now uploaded a lot of R-scripts related to lectures and exercises. Please give me a notice on geirs@math.uio.no if I have forgot some.
There is one compulsory exercise in this course. The exercise will be made available mid October with a deadline end of October.
It will be a written exam similar to earlier exams (although the 2020 exam is a bit special since it was a home exam). The only permitted aid will be an approved calculator, but formulas for distribution functions and such will be provided as part of the exam set.
For PhD students, an additional oral presentation is required. In principle this should be performed before the exam, but due to that the exam is very early this year, we will do this after the exam. There will be two options regarding this presentation:
- Either give a 15 minutes presentation on how Bayesian methods can be applied or useful in your own research
- Or you give a 15 minutes presentation on an existing paper about Bay...
Amir and Anders have kindly volunteered to be student representatives in the course.
Within a few weeks we will have a course evaluation where I will leave the lecture and you can discuss how the course have been so far and points for improvements. I will thereafter have a meeting with Amir and Anders to discuss possible actions.
It you before that have any issues related to the course that you do not want to take directly with me, you might contact the student representatives instead.
Geir
Slides for the first lecture is available under the schedule link.
There will be no recording of the lectures, but slides with lecture notes will be provided (preferable a bit before the lectures).
Geir
The first lecture will be Monday August 22. See schedule for a preliminary plan for the course. Some lecture notes will also be put out here. See the Syllabus for information about textbook. Note that the book is also available at the web, see preliminary syllabus
The course will be a mix between theory and applications.
There will be weekly exercises, both theoretical and practical. We will typically use 1 hour on Mondays to discuss (some of) these exercises.
Geir