|
wolffd@0
|
1 function [engine,engine2] = test_ho_inf_enginge(order,T)
|
|
wolffd@0
|
2
|
|
wolffd@0
|
3 assert(order >= 1)
|
|
wolffd@0
|
4 % Model a SISO system, i. e. all node are one-dimensional
|
|
wolffd@0
|
5 % The nodes are numbered as follows
|
|
wolffd@0
|
6 % u(t) = 1 input
|
|
wolffd@0
|
7 % y(t) = 2 model output
|
|
wolffd@0
|
8 % z(t) = 3 noise
|
|
wolffd@0
|
9 % q(t) = 4 observed output = noise + model output
|
|
wolffd@0
|
10
|
|
wolffd@0
|
11 ns = [1 1 1 1];
|
|
wolffd@0
|
12
|
|
wolffd@0
|
13 % Model a linear system, i.e. there are no discrete nodes
|
|
wolffd@0
|
14 dn = [];
|
|
wolffd@0
|
15
|
|
wolffd@0
|
16 % Modeling of connections within a time slice
|
|
wolffd@0
|
17 intra = zeros(4);
|
|
wolffd@0
|
18 intra(2,4) = 1; % Connection y(t) -> q(t)
|
|
wolffd@0
|
19 intra(3,4) = 1; % Connection z(t) -> q(t)
|
|
wolffd@0
|
20
|
|
wolffd@0
|
21 % Connections to the next time slice
|
|
wolffd@0
|
22 inter = zeros(4,4,order);
|
|
wolffd@0
|
23 inter(1,2,1) = 1; % u(t) -> y(t+1);
|
|
wolffd@0
|
24 inter(2,2,1) = 1; %y(t) -> y(t+1);
|
|
wolffd@0
|
25 inter(3,3,1) = 1; %z(t) -> z(t+1);
|
|
wolffd@0
|
26
|
|
wolffd@0
|
27 if order >= 2
|
|
wolffd@0
|
28 inter(1,2,2) = 1; % u(t) -> y(t+2);
|
|
wolffd@0
|
29 inter(2,2,2) = 1; % y(t) -> y(t+2);
|
|
wolffd@0
|
30 end
|
|
wolffd@0
|
31
|
|
wolffd@0
|
32 for i = 3: order
|
|
wolffd@0
|
33 inter(:,:,i) = inter(:,:,i-1); %u(t) -> y(t+i) y(t) -> y(t) +i
|
|
wolffd@0
|
34 end;
|
|
wolffd@0
|
35
|
|
wolffd@0
|
36
|
|
wolffd@0
|
37 % Compution of a higer order Markov Model
|
|
wolffd@0
|
38 bnet = mk_higher_order_dbn(intra,inter,ns,'discrete',dn);
|
|
wolffd@0
|
39 bnet2 = mk_dbn(intra,inter(:,:,1),ns,'discrete',dn)
|
|
wolffd@0
|
40
|
|
wolffd@0
|
41
|
|
wolffd@0
|
42 %Calculation of the number of nodes with different parameters
|
|
wolffd@0
|
43 %There is one input and one output nodes 2
|
|
wolffd@0
|
44 %There are two different disturbance node 2
|
|
wolffd@0
|
45 %There are order +1 nodes for y 1 + order
|
|
wolffd@0
|
46 numOfNodes = 5 + order;
|
|
wolffd@0
|
47
|
|
wolffd@0
|
48 % First input node
|
|
wolffd@0
|
49 bnet.CPD{1} = gaussian_CPD(bnet,1,'mean',0);
|
|
wolffd@0
|
50 bnet2.CPD{1} = gaussian_CPD(bnet,1,'mean',0);
|
|
wolffd@0
|
51 % Modeled output
|
|
wolffd@0
|
52 bnet.CPD{2} = gaussian_CPD(bnet,2,'mean',0);
|
|
wolffd@0
|
53 bnet2.CPD{2} = gaussian_CPD(bnet,2,'mean',0);
|
|
wolffd@0
|
54 %Disturbance
|
|
wolffd@0
|
55 bnet.CPD{3} = gaussian_CPD(bnet,3,'mean',0);
|
|
wolffd@0
|
56 bnet2.CPD{3} = gaussian_CPD(bnet,3,'mean',0);
|
|
wolffd@0
|
57
|
|
wolffd@0
|
58 %Qutput
|
|
wolffd@0
|
59 bnet.CPD{4} = gaussian_CPD(bnet,4,'mean',0);
|
|
wolffd@0
|
60 bnet2.CPD{4} = gaussian_CPD(bnet,4,'mean',0);
|
|
wolffd@0
|
61
|
|
wolffd@0
|
62
|
|
wolffd@0
|
63 %Output node in the second time-slice
|
|
wolffd@0
|
64 %Remember that node number 6 is an example for
|
|
wolffd@0
|
65 %the fifth equivalence class
|
|
wolffd@0
|
66 bnet.CPD{5} = gaussian_CPD(bnet,6,'mean',0);
|
|
wolffd@0
|
67 bnet2.CPD{5} = gaussian_CPD(bnet,6,'mean',0);
|
|
wolffd@0
|
68
|
|
wolffd@0
|
69 %Disturbance node in the second time slice
|
|
wolffd@0
|
70 bnet.CPD{6} = gaussian_CPD(bnet,7,'mean',0);
|
|
wolffd@0
|
71 bnet2.CPD{6} = gaussian_CPD(bnet,7,'mean',0);
|
|
wolffd@0
|
72
|
|
wolffd@0
|
73 % Modeling of the remaining nodes for y
|
|
wolffd@0
|
74 for i = 7:numOfNodes
|
|
wolffd@0
|
75 bnet.CPD{i} = gaussian_CPD(bnet,(i - 6)*4 + 7,'mean',0);
|
|
wolffd@0
|
76 end
|
|
wolffd@0
|
77
|
|
wolffd@0
|
78 % Generation of the inference engine
|
|
wolffd@0
|
79 engine = dv_unrolled_dbn_inf_engine(bnet,T);
|
|
wolffd@0
|
80 engine2 = jtree_unrolled_dbn_inf_engine(bnet,T);
|
|
wolffd@0
|
81
|
|
wolffd@0
|
82
|
|
wolffd@0
|
83
|
|
wolffd@0
|
84
|
|
wolffd@0
|
85
|
|
wolffd@0
|
86
|
|
wolffd@0
|
87
|
