1 | #include <itpp/itbase.h> |
---|
2 | #include "merger.h" |
---|
3 | #include "arx.h" |
---|
4 | |
---|
5 | vec merger::lognorm_merge ( mat &lW ) { |
---|
6 | int nu=lW.rows(); |
---|
7 | vec mu = sum ( lW ) /nu; //mean of logs |
---|
8 | vec lam = sum ( pow ( lW,2 ) )-nu*pow ( mu,2 ); |
---|
9 | double coef=0.0; |
---|
10 | switch ( nu ) { |
---|
11 | case 2: |
---|
12 | coef=sqrt ( beta*2 ) * ( 1-0.5*sqrt ( ( 4*beta-3 ) /beta ) ); |
---|
13 | return exp ( coef*sqrt ( lam ) + mu ); |
---|
14 | break; |
---|
15 | case 3://Ration of Bessel |
---|
16 | break; |
---|
17 | case 4: |
---|
18 | break; |
---|
19 | default: // Approximate conditional density |
---|
20 | break; |
---|
21 | } |
---|
22 | return vec ( 0 ); |
---|
23 | } |
---|
24 | |
---|
25 | void merger::merge ( const epdf* g0 ) { |
---|
26 | it_file dbg ( "merger_debug.it" ); |
---|
27 | |
---|
28 | it_assert_debug ( rv.equal ( g0->_rv() ),"Incompatible g0" ); |
---|
29 | //Empirical density - samples |
---|
30 | eEmp eSmp ( rv,Ns ); |
---|
31 | eSmp.set_parameters ( ones ( Ns ), g0 ); |
---|
32 | Array<vec> &Smp = eSmp._samples(); //aux |
---|
33 | vec &w = eSmp._w(); //aux |
---|
34 | |
---|
35 | mat Smp_ex =ones ( rv.count() +1,Ns ); // Extended samples for the ARX model - the last row is ones |
---|
36 | for ( int i=0;i<Ns;i++ ) { set_col_part ( Smp_ex,i,Smp ( i ) );} |
---|
37 | |
---|
38 | dbg << Name ( "Smp_0" ) << Smp_ex; |
---|
39 | |
---|
40 | // Stuff for merging |
---|
41 | vec lw_src ( Ns ); |
---|
42 | vec lw_mix ( Ns ); |
---|
43 | mat lW=zeros ( n,Ns ); |
---|
44 | vec vec0 ( 0 ); |
---|
45 | |
---|
46 | // Initial component in the mixture model |
---|
47 | mat V0=1e-8*eye ( rv.count() +1 ); |
---|
48 | ARX A0 ( RV ( "{th_r }", vec_1 ( rv.count() * ( rv.count() +1 ) ) ),\ |
---|
49 | V0, rv.count() *rv.count() +5.0 ); //initial guess of Mix: zero mean, large variance |
---|
50 | |
---|
51 | Mix.init ( &A0, Smp_ex, Nc ); |
---|
52 | //Preserve initial mixture for repetitive estimation via flattening |
---|
53 | MixEF Mix_init(Mix); |
---|
54 | |
---|
55 | // ============= MAIN LOOP ================== |
---|
56 | bool converged=false; |
---|
57 | int niter = 0; |
---|
58 | char str[100]; |
---|
59 | |
---|
60 | epdf* Mpred; |
---|
61 | vec Mix_pdf ( Ns ); |
---|
62 | while ( !converged ) { |
---|
63 | //Re-estimate Mix |
---|
64 | //Re-Initialize Mixture model |
---|
65 | Mix.flatten(&Mix_init); |
---|
66 | Mix.bayesB ( Smp_ex, w*Ns ); |
---|
67 | Mpred = Mix.predictor ( rv ); // Allocation => must be deleted at the end!! |
---|
68 | |
---|
69 | sprintf ( str,"Mpred_mean%d",niter ); |
---|
70 | dbg << Name ( str ) << Mpred->mean(); |
---|
71 | |
---|
72 | // Generate new samples |
---|
73 | eSmp.set_samples ( Mpred ); |
---|
74 | for ( int i=0;i<Ns;i++ ) { set_col_part ( Smp_ex,i,Smp ( i ) );} |
---|
75 | |
---|
76 | sprintf ( str,"Mpdf%d",niter ); |
---|
77 | for ( int i=0;i<Ns;i++ ) {Mix_pdf ( i ) = Mix.logpred ( Smp_ex.get_col ( i ) );} |
---|
78 | dbg << Name ( str ) << Mix_pdf; |
---|
79 | |
---|
80 | sprintf ( str,"Smp%d",niter ); |
---|
81 | dbg << Name ( str ) << Smp_ex; |
---|
82 | |
---|
83 | //Importace weighting |
---|
84 | for ( int i=0;i<n;i++ ) { |
---|
85 | lw_src=0.0; |
---|
86 | //======== Same RVs =========== |
---|
87 | //Split according to dependency in rvs |
---|
88 | if ( mpdfs ( i )->_rv().count() ==rv.count() ) { |
---|
89 | // no need for conditioning or marginalization |
---|
90 | for ( int j=0;j<Ns; j++ ) { // Smp is Array<> => for cycle |
---|
91 | lw_src ( j ) =mpdfs ( i )->_epdf().evalpdflog ( Smp ( j ) ); |
---|
92 | } |
---|
93 | } |
---|
94 | else { |
---|
95 | // compute likelihood of marginal on the conditional variable |
---|
96 | if ( mpdfs ( i )->_rvc().count() >0 ) { |
---|
97 | // Make marginal on rvc_i |
---|
98 | epdf* tmp_marg = Mpred->marginal ( mpdfs ( i )->_rvc() ); |
---|
99 | //compute vector of lw_src |
---|
100 | for ( int k=0;k<Ns;k++ ) { |
---|
101 | lw_src ( k ) += tmp_marg->evalpdflog ( dls ( i )->get_val ( Smp ( i ) ) ); |
---|
102 | } |
---|
103 | delete tmp_marg; |
---|
104 | } |
---|
105 | // Compute likelihood of the missing variable |
---|
106 | if ( rv.count() > ( mpdfs ( i )->_rv().count() + mpdfs ( i )->_rvc().count() ) ) { |
---|
107 | /////////////// |
---|
108 | // There are variales unknown to mpdfs(i) : rvzs |
---|
109 | mpdf* tmp_cond = Mpred->condition ( rvzs ( i ) ); |
---|
110 | // Compute likelihood |
---|
111 | vec lw_dbg=lw_src; |
---|
112 | for ( int k= 0; k<Ns; k++ ) { |
---|
113 | lw_src ( k ) += log ( |
---|
114 | tmp_cond->evalcond ( |
---|
115 | zdls ( i )->get_val ( Smp ( k ) ), |
---|
116 | zdls ( i )->get_cond ( Smp ( k ) ) ) ); |
---|
117 | } |
---|
118 | delete tmp_cond; |
---|
119 | } |
---|
120 | // Compute likelihood of the partial source |
---|
121 | for ( int k= 0; k<Ns; k++ ) { |
---|
122 | mpdfs ( i )->condition ( dls ( i )->get_cond ( Smp ( k ) ) ); |
---|
123 | lw_src ( k ) += mpdfs ( i )->_epdf().evalpdflog ( dls ( i )->get_val ( Smp ( k ) ) ); |
---|
124 | } |
---|
125 | |
---|
126 | } |
---|
127 | lW.set_row ( i, lw_src ); // do not divide by mix |
---|
128 | } |
---|
129 | //Importance of the mixture |
---|
130 | for ( int j=0;j<Ns;j++ ) { |
---|
131 | lw_mix ( j ) =Mix.logpred ( Smp_ex.get_col ( j ) ); |
---|
132 | } |
---|
133 | sprintf ( str,"lW%d",niter ); |
---|
134 | dbg << Name ( str ) << lW; |
---|
135 | |
---|
136 | w = lognorm_merge ( lW ); //merge |
---|
137 | |
---|
138 | sprintf ( str,"w%d",niter ); |
---|
139 | dbg << Name ( str ) << w; |
---|
140 | sprintf ( str,"lw_m%d",niter ); |
---|
141 | dbg << Name ( str ) << lw_mix; |
---|
142 | |
---|
143 | //Importance weighting |
---|
144 | w /=exp ( lw_mix ); // hoping that it is not numerically sensitive... |
---|
145 | //renormalize |
---|
146 | w /=sum ( w ); |
---|
147 | |
---|
148 | sprintf ( str,"w_is_%d",niter ); |
---|
149 | dbg << Name ( str ) << w; |
---|
150 | |
---|
151 | // eSmp.resample(); // So that it can be used in bayes |
---|
152 | // for ( int i=0;i<Ns;i++ ) { set_col_part ( Smp_ex,i,Smp ( i ) );} |
---|
153 | |
---|
154 | sprintf ( str,"Smp_res%d",niter ); |
---|
155 | dbg << Name ( str ) << Smp; |
---|
156 | |
---|
157 | // ==== stopping rule === |
---|
158 | niter++; |
---|
159 | converged = ( niter>9 ); |
---|
160 | } |
---|
161 | |
---|
162 | } |
---|