1 | function outIndex = resample_residual(inIndex,q); |
---|
2 | % PURPOSE : Performs the resampling stage of the SIR |
---|
3 | % in order(number of samples) steps. It uses Liu's |
---|
4 | % residual resampling algorithm and Niclas' magic line. |
---|
5 | % INPUTS : - inIndex = Input particle indices. |
---|
6 | % - q = Normalised importance ratios. |
---|
7 | % OUTPUTS : - outIndex = Resampled indices. |
---|
8 | |
---|
9 | % AUTHORS : Arnaud Doucet, Nando de Freitas and Neil Gordon |
---|
10 | % DATE : 08-09-98 |
---|
11 | |
---|
12 | if nargin < 2, error('Not enough input arguments.'); end |
---|
13 | |
---|
14 | [S,arb] = size(q); % S = Number of particles. |
---|
15 | |
---|
16 | % RESIDUAL RESAMPLING: |
---|
17 | % ==================== |
---|
18 | N_babies= zeros(1,S); |
---|
19 | % first integer part |
---|
20 | q_res = S.*q'; |
---|
21 | N_babies = fix(q_res); |
---|
22 | % residual number of particles to sample |
---|
23 | N_res=S-sum(N_babies); |
---|
24 | if (N_res~=0) |
---|
25 | q_res=(q_res-N_babies)/N_res; |
---|
26 | cumDist= cumsum(q_res); |
---|
27 | % generate N_res ordered random variables uniformly distributed in [0,1] |
---|
28 | u = fliplr(cumprod(rand(1,N_res).^(1./(N_res:-1:1)))); |
---|
29 | j=1; |
---|
30 | for i=1:N_res |
---|
31 | while (u(1,i)>cumDist(1,j)) |
---|
32 | j=j+1; |
---|
33 | end |
---|
34 | N_babies(1,j)=N_babies(1,j)+1; |
---|
35 | end; |
---|
36 | end; |
---|
37 | |
---|
38 | % COPY RESAMPLED TRAJECTORIES: |
---|
39 | % ============================ |
---|
40 | index=1; |
---|
41 | for i=1:S |
---|
42 | if (N_babies(1,i)>0) |
---|
43 | for j=index:index+N_babies(1,i)-1 |
---|
44 | outIndex(j) = inIndex(i); |
---|
45 | end; |
---|
46 | end; |
---|
47 | index= index+N_babies(1,i); |
---|
48 | end |
---|
49 | |
---|
50 | |
---|
51 | |
---|
52 | |
---|
53 | |
---|
54 | |
---|
55 | |
---|
56 | |
---|
57 | |
---|
58 | |
---|
59 | |
---|
60 | |
---|