02.09.2014 Views

multivariate poisson hidden markov models for analysis of spatial ...

multivariate poisson hidden markov models for analysis of spatial ...

multivariate poisson hidden markov models for analysis of spatial ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

m<br />

∑<br />

α ( i)<br />

= α ( i −1)<br />

P f ( y ; λ )<br />

(5.27)<br />

j<br />

k = 1<br />

k<br />

(1)<br />

[ α 1) = P f ( y ; λ ), j 1,..., m]<br />

, and<br />

j<br />

(<br />

j 1 j<br />

=<br />

m<br />

∑<br />

k = 1<br />

kj<br />

i+ 1<br />

; λ<br />

k<br />

)<br />

k<br />

( i + 1)<br />

i<br />

j<br />

β ( i)<br />

= P f ( y β<br />

(5.28)<br />

j<br />

jk<br />

[ β ( n)<br />

= 1, j 1,..., m]<br />

.<br />

j<br />

=<br />

Note that the α'<br />

s are computed by a <strong>for</strong>ward pass through the observations and the β ' s<br />

by a backward pass after evaluating the α'<br />

s. The likelihood is then simply calculated by<br />

the expression∑ α<br />

j<br />

( n)<br />

.<br />

m<br />

j=<br />

1<br />

The calculations <strong>of</strong> X<br />

1,<br />

X<br />

2,<br />

X<br />

3,<br />

X<br />

12,<br />

X<br />

13,<br />

and X 23<br />

can be carried out using the same<br />

<strong>for</strong>mulas explained in section 5.2.2. The <strong>multivariate</strong> Poisson model is defined as<br />

Y = X + X + X<br />

1 1 12 13<br />

Y = X + X + X<br />

2 2 12 23<br />

Y = X + X + X<br />

3 3 13 23<br />

.<br />

(5.29)<br />

E-step: Using the current values <strong>of</strong> the parameters calculate<br />

E[ X | Y, u ( i) = 1, Φ ] = d<br />

j<br />

j<br />

12i j 12i<br />

min( y1i, y2i)<br />

rPo( y1 i<br />

−r | λ1j) Po( y2i −r | λ2j) Po( r | λ12j)<br />

= ∑ . (5.30)<br />

f ( y | λ )<br />

r=<br />

0<br />

i<br />

j<br />

j<br />

j<br />

The corresponding expressions <strong>for</strong> E[ X13 i<br />

| Y, uj( i) = 1, Φ ] = d13<br />

i<br />

and<br />

E[ X | Y, u ( i) = 1, Φ ] = d follow by analogy. Then<br />

j<br />

j<br />

23i j 23i<br />

96

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!