Information Technology Reference
In-Depth Information
parameters correspond to the parameters in each distributions (i.e. shape , scale
in gamma and mean , standard deviation in normal and lognormal distributions).
The detail of the EM algorithm is as follows:
1. Initialize θ A and θ B . It is done by first randomly partitioning the dataset
into K groups and then calculate method of moment estimates for each of
the groups.
2. M-step: Given p ( x|j ) , maximize loglikelihood (LL) with respect to the pa-
rameters θ A
and θ B . We obtain maximizer for [ θ A 1 , θ B 1 ,...,θ A K , θ B K ]nu-
merically.
3. E-step: Given the parameter estimates from M-step, we compute:
|x, θ A 1 , θ B 1 ,...,θ A K , θ B K ]
p ( x|j ) = E [ p ( x|j )
(4)
p ( x|j )
j =1 p ( x|j )
=
(5)
4. Repeat M-step and E-step until the change in the value of the loglikelihood
(LL) is negligible.
In order to avoid local maxima, we run the above EM algorithm ten times with
different starting points.
2.2 Model Selection
When fitting mixture models to expression data, it is necessary to desierable to
choose an appropriate number of components, which fits the data well but does
not overfit. For this task we tried two information criteria: AIC (Akaike Informa-
tion Criterion [10]) and BIC (Bayesian Information Criterion [11]). Specifically:
AIC =
2 LL +2 c
(6)
and
BIC =
2 LL + clog ( N )
(7)
To choose models, we fit mixture models with the EM algorithm for one to five
components and chose the model with the smallest information criteria value (the
degree of freedom c in the above formulas, is equal to 3 K − 1for K components).
3 Experimental Results
3.1 Estimating Number of Components
We generated simulated datasets from mixture models containing with one, two
and three components. We performed two sets of equivalent experiments, one
using the gamma and one using the lognormal distribution for the mixture model
components. For the component parameters, each distinct combination of the
 
Search WWH ::




Custom Search