Midproject review
Objective
Recall that the objective of your project is to
combine a factorial hidden Markov with a phylogenetic
tree for 4-species alignments to distinguish between
recombination and rate variation.
This project builds on the following earlier work:
And this is, in essence, what your thesis is about:
Objective: Detecting rate variation and recombination.
Model: Phylogenetic tree plus two parallel (factorial) hidden Markov model.
Learning method: Either maximum likelihood (EM algorithm)
or MCMC.
You can follow two routes:
|
Route A ("classical"):
|
Optimizing the parameters with maximum likelihood
|
|
Route B ("Bayesian"):
|
Sampling the parameters from the posterior with MCMC
|
Route A might be faster because
it seems to be more straightforward to use existing functions
implemented in
Kevin Murphy's
Bayes Net Toolbox
and
Hidden Markov Model (HMM) Toolbox.
It is your task to familiarize yourself with these software
packages and to decide which functions are useful for your purposes,
and how to use them.
Route B is, in principle, the more powerful method;
and it has the advantage of leading to some
new intermediate results that you can include in your thesis.
In drafting the next milestone sections, I will assume that you
follow route B. Feel free to explore route A yourself.
Current situation
You should by now have obtained a
sufficiently profound background
in probabilistic modelling and learning
from data,
obtained by
-
attending the courses
LFD1
and
PMR;
-
doing the previous
MSc project;
-
and doing the exercises for the previous milestones.
This should, in principle, put you in the
position to
-
work much more independently;
-
explore various options yourself and assess them critically;
-
derive new inference algorithms yourself;
-
implement them in software and verify the correctness
of your implementation.
Make sure that you keep focused on these objectives;
you have not quite got there yet.
Additional help
You may use the program packages
SERAD and
BARCE
for comparison with your own software implementation.
To download SERAD, click here.
Back to the previous page.