Messages
In exercise 8 the distribution for x_i included an y_i which shouldn't have been there. Further, there was an extra mu_3 there. This has been corrected.
Exercise 5) a beta_0 term was missing in the likelihood expression. This is now corrected.
There was a missing term on the H-function. This is now corrected.
All students should now have got feedback on the compulsory exercise.
There are some that needs to resubmit their exercise. Note that the deadline for resubmission is already May 9. Note further that it is allowed to come and ask for help!
On the presentation for the EM algorithm in the M step for the HMM model there was a mistake in the formulae for updating the variance (index t on mu should be index t+1). This is now corrected.
Regarding the estimation of the variance: Many students seems to get quite large estimates here (around 100-150). This is actually ok, because when I simulated the data, I used a standard deviation of 10 and not sqrt(10). Sorry for this!
Point g has been changed somewhat to make the calculations a bit simpler.
In exercise 1 e) you might obtain numerical problems with the prior Gamma(0.01, 0.01). If so, try out Gamma(0.1, 0.1) instead.
In part 1, there was a small typo in the definition of the model in Ex 1, were the index i-1 should have been t-1. This is now corrected.
The second part of the compulsory exercise is now published. The second exercise is about the MCMC algorithm which will be the topic for the lectures in the next weeks.
A new version of the presentation for ch 4 is put out which includes a page for calculation of likelihoods in HMM's. Can be used for the computsory exercise.
The smc_cos.R routine had some errors. This is now corrected. Thanks to Lars Henry!
There was a misprint in extra exercise 7 (wrong sign on a term) that now has been corrected.
If you find other typos, please tell me!
Geir
The lectures are now moved to UE108, same building, but first floor.
The first lecture will, through several examples, illustrate what kind of computational problems that occur in statistics/data analysis/machine learning. An overview of the type of methods that will be covered will be discussed as well.
If time, we will start on chapter two of the text book (see Syllabus).
I will mostly use presentations on the screen, but sometimes also the blackboard. Presentations are/will be available under Schedule.
We will typically use 2 hours for lecturing and 1 hour for going through exercises.
Some exercises requires use of the computer. I will mainly use R, but other options are possible.
Geir