Messages

Thanks for your considerable efforts regarding the Exam Project, alias Exam Part One. Now for Exam Part Two, the four-hour written exam on Wednesday December 16, 9-13 (where you ought to find the required geographical information via StudentWeb etc.). As agreed, this is a *no book* exam, but each student is allowed to bring along *precisely one sheet of handwritten notes* (one side, not both sides). Also, please bring an ordinary "pocket calculator" (no noise, no memory, no intergallactic communication capabilities, no Metropolis-Hastings). No "heavy computation" will be required. Good luck.

There's another text bit which is unclear from my side, unfortunately. For 3(c), I had "with n = 100 and with true value \tau_\true = 66" in my manuscript, but somehow the "n = 100" part disappeared; mea culpa. So please put in n = 100 if you see this. Students will get credit here as long as they do something reasonable.

I have not been careful enough regarding point 3(a). Mea culpa! The correct text there should be:

If $\tau$ is given the flat prior on $1,\ldots,n$ (each position having prior probability $1/n$), find an expression for the posterior distribution $\pi(\tau|\data)$. 

This message is also given under the "Exam Project, December 2015" umbrella.

1. We meet Friday November 27, but without a fixed list of exercises to go through. I will have a quick going-through of main issues for the course.

2. A good exam preparation exercise is to invent say five to ten things you might expect Nils to touch when he constructs his exam sets (the project + the four-four exam). You may also prepare questions from the course & its exercises, and these will be tended to as time permits.

3. Note that I have uploaded the four-hour exam set from December 2013.

4. Nils Lid Hjort and Celine Cunan give a talk Thursday November 26 at 18:04, in the "Pub med professor" series, Realistforeningen, Vilhelm Bjerknes hus. There's a slight positive empirical correlation between Nils's talks and the exam projects he constructs.

https://www.facebook.com/events/161954000825163/

As agreed last time we first do the remaining Exam stk 4021 2013 Exercise #3, along with some further discussion of details. Then work through Exam stk 4020 2008 Exercise #1, #3.

On Friday Nov 6 I went through Nils Exericses #24 and #25, along with further discussions and illustrations of empirical Bayes setups, in situations where an ensemble of parameters are to be estimated.

For Friday Nov 13, work through Exam stk 4021 2013 #2, then #1, then #3.

1. On Fri Oct 23 I went through various earlier exam exercises, including the Sicut cervus disederat at fontes aquarum from 2004, and more on MCMC, and started discussing Bayesian aspects of regression models, Ch 14.

2. Exercises for Fri Oct 30: First, exam stk 4020 2004, exercise 3, hjorter i skogen: Do the last point, with the prior on (theta1, theta2) that is constant over the area theta1 < theta2. Then, exam stk 4021 2012, exercise 2, value of cars as a function of age. Then, stk 4020 2004 again, exercise 4, the hierarchical probabilities, but this time with (k, p0) both unknown parameters, and with MCMC to handle this case. As prior, take p0 flat and k as given in the exercise.

1. On Friday Oct 16 I wore my Alan Gelfand t-shirt, wrapped in Markov chains, and discussed MCMC, with emphasis on the Metropolis-Hastings methods. I also went through various exercises.

2. Exercises for Friday Oct 23: Exam stk 4020 2004, #2 (hjorter i skogen). Then #3 from the same set, point (h), again!, but now with a MCMC put up to handle the uncertainty in both parameters (k, p0). Then Nils Exercise 19(a)-(e), 21.

3. I will also gently start the last chapter on the curriculum list, namely chapter 14, on regression models.

On Fri Oct 9 I went through some of the background for MCMC, and we're ready to learn more practicalities regarding how to set up the chains etc.

Exercises: First, Nils Exercises 18, 19(a)-(b), 17(d). Then the Poisson forest Exam stk 4020 exercise 1 from 2004.

1. On Friday I went through the essence of Ch 5, on hierarchical models. Soon we start "the last new big thing" in the course, the world of MCMC.

2. Exercises: more, from earlier exam sets! First: stk 4020, 2004, #4 (hierarchical modelling of probabilities). Then: stk 4020, 2012, #1 (Galton). Then: stk 4020, 2004, #1 (Poisson forest).

** Extra message (appended later): Due to a couple of questions during the week I decide to start Friday Oct 9 with just a few minutes doing the book's Exercise 2.4.

1. On Fri Sept 25 I went through & discussed the essentials of Ch 4. Next week we start Ch 5, and soon we'll finally enter the realm of the MCMCs. I also went through various exercises.

2. Exercises for Fri Oct 2: First we spend a few minutes with details from Nils Exercises 22(h). Then we do exercises from previous Exam Projects. First Exam stk 4020 2008, 2 (Abel); then Exam stk 4021 2012, 1 (Galton); then, if we have time, Exam stk 4020 2004, 1 (Poisson forest). This 2004 exam set can be found under stk 4021 2013. 

1. On Fri 18 Sept I went through the essential parts of the second half of Ch 3, with some emphasis on the Dirichlet-multinomial setup. Next week I start with Ch 4 material.

2. For next week, do Exercise 3.2 from the book, and Nils Exercises 9(e), 22(f)-(g), 23(h).

From time to time I will upload some "R scripts", to be used inside the R computing environment e.g. using

source ("com34a")

etc. These R scripts may then be used, copied, modifed, extended, ameliorated as the need arises. I consciously plan to *not* upload every R script I invent for the semester, as the students need to develop these programming skills themselves.

1. Next week I'm in Bruxelles, but my PhD student C¨¦line Cunen will teach in loco Hjorti. From Fri Sept 11 onwards the time window moves to 9:15 - 12:00, but for Fri Sept 4 we still stick to 10:15 - 13:00.

2. Today I went through Nils Exercises 1-2 and a few exercises from Ch 1, and used time on the Bayesian Master Recipe, cf. Nils Exercise 2. Next week C¨¦line will lecture from the start of Ch 2, in addition to going through exercises.

3. I'm now listing some exercises, where C¨¦line will go through some on Sept 4 and I the rest on Sept 11. First, some more details from the book's 1.1; then, Nils Exercise 12; then the book's 2.1, 2.3, 2.4, 2.8, 2.10, 2.12, 2.16.

Welcome to stk 4021 / 9021, Applied Bayesian Analysis and Numerical Methods! Lectures take place Fridays 10-13, in Undervisningsrom UE26 ("underetasjen" in the Mathematics Building), starting August 21.

The course material will be based on (1) some of the chapters from "Bayesian Data Analysis" (Gelman, Carlin, Stern, Dunson, Vehtari, Rubin, *3rd edition*, 2014); (2) certain "Exercises and Lecture Notes" by Nils Lid Hjort (to be made available at this site); and (3) some of the course notes made by Thordis Thorarinsdottir when she taught this course autumn 2014 (check the course website).

Copies of the Gelman et al. (2014) book have been ordered (as of mid June 2015) for the Akademika bookstore, but it might be as easy to order it via Amazon, eBay, CRC Press, or yet other links.

Nils Lid Hjort