1 | #include "arx.h" |
---|
2 | |
---|
3 | using namespace itpp; |
---|
4 | |
---|
5 | void ARX::bayes ( const vec &dt, const double w ) { |
---|
6 | double lnc; |
---|
7 | |
---|
8 | if ( frg<1.0 ) { |
---|
9 | est.pow ( frg ); |
---|
10 | if ( evalll ) { |
---|
11 | last_lognc = est.lognc(); |
---|
12 | } |
---|
13 | } |
---|
14 | V.opupdt ( dt,w ); |
---|
15 | nu+=w; |
---|
16 | |
---|
17 | if ( evalll ) { |
---|
18 | lnc = est.lognc(); |
---|
19 | ll = lnc - last_lognc; |
---|
20 | last_lognc = lnc; |
---|
21 | } |
---|
22 | } |
---|
23 | |
---|
24 | double ARX::logpred ( const vec &dt ) const { |
---|
25 | egiw pred ( est ); |
---|
26 | ldmat &V=pred._V(); |
---|
27 | double &nu=pred._nu(); |
---|
28 | |
---|
29 | double lll; |
---|
30 | |
---|
31 | if ( frg<1.0 ) { |
---|
32 | pred.pow ( frg ); |
---|
33 | lll = pred.lognc(); |
---|
34 | } |
---|
35 | else |
---|
36 | if(evalll){lll=last_lognc;}else{lll=pred.lognc();} |
---|
37 | |
---|
38 | V.opupdt ( dt,1.0 ); |
---|
39 | nu+=1.0; |
---|
40 | |
---|
41 | return pred.lognc()-lll; |
---|
42 | } |
---|
43 | |
---|
44 | BM* ARX::_copy_ ( bool changerv) { |
---|
45 | ARX* Tmp=new ARX ( *this ); |
---|
46 | if ( changerv ) {Tmp->rv.newids(); Tmp->est._renewrv(Tmp->rv);} |
---|
47 | return Tmp; |
---|
48 | } |
---|
49 | |
---|
50 | void ARX::set_statistics ( const BMEF* B0 ) { |
---|
51 | const ARX* A0=dynamic_cast<const ARX*> ( B0 ); |
---|
52 | |
---|
53 | it_assert_debug ( V.rows() ==A0->V.rows(),"ARX::set_statistics Statistics differ" ); |
---|
54 | set_parameters ( A0->V,A0->nu ); |
---|
55 | } |
---|
56 | /*! \brief Return the best structure |
---|
57 | @param Eg a copy of GiW density that is being examined |
---|
58 | @param Eg0 a copy of prior GiW density before estimation |
---|
59 | @param Egll likelihood of the current Eg |
---|
60 | @param indeces current indeces |
---|
61 | \return best likelihood in the structure below the given one |
---|
62 | */ |
---|
63 | double egiw_bestbelow ( egiw Eg, egiw Eg0, double Egll, ivec &indeces ) { //parameter Eg is a copy! |
---|
64 | ldmat Vo = Eg._V(); //copy |
---|
65 | ldmat Vo0 = Eg._V(); //copy |
---|
66 | ldmat& Vp = Eg._V(); // pointer into Eg |
---|
67 | ldmat& Vp0 = Eg._V(); // pointer into Eg |
---|
68 | int end = Vp.rows()-1; |
---|
69 | int i; |
---|
70 | mat Li; |
---|
71 | mat Li0; |
---|
72 | double maxll=Egll; |
---|
73 | double tmpll=Egll; |
---|
74 | double belll=Egll; |
---|
75 | |
---|
76 | ivec tmpindeces; |
---|
77 | ivec maxindeces=indeces; |
---|
78 | |
---|
79 | |
---|
80 | cout << "bb:(" << indeces <<") ll=" << Egll <<endl; |
---|
81 | |
---|
82 | //try to remove only one rv |
---|
83 | for ( i=0;i<end;i++ ) { |
---|
84 | //copy original |
---|
85 | Li = Vo._L(); |
---|
86 | Li0 = Vo0._L(); |
---|
87 | //remove stuff |
---|
88 | Li.del_col ( i+1 ); |
---|
89 | Li0.del_col ( i+1 ); |
---|
90 | Vp.ldform ( Li,Vo._D() ); |
---|
91 | Vp0.ldform ( Li0,Vo0._D() ); |
---|
92 | tmpll = Eg.lognc()-Eg0.lognc(); // likelihood is difference of norm. coefs. |
---|
93 | |
---|
94 | cout << "i=(" << i <<") ll=" << tmpll <<endl; |
---|
95 | |
---|
96 | // |
---|
97 | if ( tmpll > Egll ) { //increase of the likelihood |
---|
98 | tmpindeces = indeces; |
---|
99 | tmpindeces.del ( i ); |
---|
100 | //search for a better match in this substructure |
---|
101 | belll=egiw_bestbelow ( Eg, Eg0, tmpll, tmpindeces ); |
---|
102 | if ( belll>maxll ) { //better match found |
---|
103 | maxll = belll; |
---|
104 | maxindeces = tmpindeces; |
---|
105 | } |
---|
106 | } |
---|
107 | } |
---|
108 | indeces = maxindeces; |
---|
109 | return maxll; |
---|
110 | } |
---|
111 | |
---|
112 | ivec ARX::structure_est ( egiw est0 ) { |
---|
113 | ivec ind=linspace ( 1,rv.count()-1 ); |
---|
114 | egiw_bestbelow ( est, est0, est.lognc()- est0.lognc(), ind ); |
---|
115 | return ind; |
---|
116 | } |
---|