1 | /*! |
---|
2 | \file |
---|
3 | \brief Bayesian Filtering using stochastic sampling (Particle Filters) |
---|
4 | \author Vaclav Smidl. |
---|
5 | |
---|
6 | ----------------------------------- |
---|
7 | BDM++ - C++ library for Bayesian Decision Making under Uncertainty |
---|
8 | |
---|
9 | Using IT++ for numerical operations |
---|
10 | ----------------------------------- |
---|
11 | */ |
---|
12 | |
---|
13 | #ifndef PARTICLES_H |
---|
14 | #define PARTICLES_H |
---|
15 | |
---|
16 | |
---|
17 | #include "../estim/arx_ext.h" |
---|
18 | #include "../stat/emix.h" |
---|
19 | |
---|
20 | namespace bdm { |
---|
21 | |
---|
22 | //! class used in PF |
---|
23 | class MarginalizedParticle : public BM{ |
---|
24 | protected: |
---|
25 | //! discrte particle |
---|
26 | dirac est_emp; |
---|
27 | //! internal Bayes Model |
---|
28 | shared_ptr<BM> bm; |
---|
29 | //! pdf with for transitional par |
---|
30 | shared_ptr<pdf> par; // pdf for non-linear part |
---|
31 | //! link from this to bm |
---|
32 | shared_ptr<datalink_part> cond2bm; |
---|
33 | //! link from cond to par |
---|
34 | shared_ptr<datalink_part> cond2par; |
---|
35 | //! link from emp 2 par |
---|
36 | shared_ptr<datalink_part> emp2bm; |
---|
37 | //! link from emp 2 par |
---|
38 | shared_ptr<datalink_part> emp2par; |
---|
39 | |
---|
40 | //! custom posterior - product |
---|
41 | class eprod_2:public eprod_base { |
---|
42 | protected: |
---|
43 | MarginalizedParticle ∓ |
---|
44 | public: |
---|
45 | eprod_2(MarginalizedParticle &m):mp(m){} |
---|
46 | const epdf* factor(int i) const {return (i==0) ? &mp.bm->posterior() : &mp.est_emp;} |
---|
47 | const int no_factors() const {return 2;} |
---|
48 | } est; |
---|
49 | |
---|
50 | public: |
---|
51 | MarginalizedParticle():est(*this){}; |
---|
52 | MarginalizedParticle(const MarginalizedParticle &m2):est(*this){ |
---|
53 | bm = m2.bm->_copy(); |
---|
54 | est_emp = m2.est_emp; |
---|
55 | par = m2.par; |
---|
56 | validate(); |
---|
57 | }; |
---|
58 | BM* _copy() const{return new MarginalizedParticle(*this);}; |
---|
59 | void bayes(const vec &dt, const vec &cond){ |
---|
60 | vec par_cond(par->dimensionc()); |
---|
61 | cond2par->filldown(cond,par_cond); // copy ut |
---|
62 | emp2par->filldown(est_emp._point(),par_cond); // copy xt-1 |
---|
63 | |
---|
64 | //sample new particle |
---|
65 | est_emp.set_point(par->samplecond(par_cond)); |
---|
66 | //if (evalll) |
---|
67 | vec bm_cond(bm->dimensionc()); |
---|
68 | cond2bm->filldown(cond, bm_cond);// set e.g. ut |
---|
69 | emp2bm->filldown(est_emp._point(), bm_cond);// set e.g. ut |
---|
70 | bm->bayes(dt, bm_cond); |
---|
71 | ll=bm->_ll(); |
---|
72 | } |
---|
73 | const eprod_2& posterior() const {return est;} |
---|
74 | |
---|
75 | void set_prior(const epdf *pdf0){ |
---|
76 | const eprod *ep=dynamic_cast<const eprod*>(pdf0); |
---|
77 | if (ep){ // full prior |
---|
78 | bdm_assert(ep->no_factors()==2,"Incompatible prod"); |
---|
79 | bm->set_prior(ep->factor(0)); |
---|
80 | est_emp.set_point(ep->factor(1)->sample()); |
---|
81 | } else { |
---|
82 | // assume prior is only for emp; |
---|
83 | est_emp.set_point(pdf0->sample()); |
---|
84 | } |
---|
85 | } |
---|
86 | |
---|
87 | /*! parse structure |
---|
88 | \code |
---|
89 | class = "BootstrapParticle"; |
---|
90 | parameter_pdf = {class = 'epdf_offspring', ...}; |
---|
91 | observation_pdf = {class = 'epdf_offspring',...}; |
---|
92 | \endcode |
---|
93 | If rvs are set, then it checks for compatibility. |
---|
94 | */ |
---|
95 | void from_setting(const Setting &set){ |
---|
96 | BM::from_setting ( set ); |
---|
97 | par = UI::build<pdf> ( set, "parameter_pdf", UI::compulsory ); |
---|
98 | bm = UI::build<BM> ( set, "bm", UI::compulsory ); |
---|
99 | } |
---|
100 | |
---|
101 | void to_setting(const Setting &set){ |
---|
102 | if (BM::log_level[logfull]){ |
---|
103 | } |
---|
104 | } |
---|
105 | void validate(){ |
---|
106 | if (est_emp.point.length()!=par->dimension()) |
---|
107 | est_emp.set_point(zeros(par->dimension())); |
---|
108 | est.validate(); |
---|
109 | |
---|
110 | yrv = bm->_yrv(); |
---|
111 | dimy = bm->dimensiony(); |
---|
112 | set_rv( concat(bm->_rv(), par->_rv())); |
---|
113 | set_dim( par->dimension()+bm->dimension()); |
---|
114 | |
---|
115 | rvc = par->_rvc(); |
---|
116 | rvc.add(bm->_rvc()); |
---|
117 | rvc=rvc.subt(par->_rv()); |
---|
118 | rvc=rvc.subt(par->_rv().copy_t(-1)); |
---|
119 | rvc=rvc.subt(bm->_rv().copy_t(-1)); // |
---|
120 | |
---|
121 | cond2bm=new datalink_part; |
---|
122 | cond2par=new datalink_part; |
---|
123 | emp2bm =new datalink_part; |
---|
124 | emp2par =new datalink_part; |
---|
125 | cond2bm->set_connection(bm->_rvc(), rvc); |
---|
126 | cond2par->set_connection(par->_rvc(), rvc); |
---|
127 | emp2bm->set_connection(bm->_rvc(), par->_rv()); |
---|
128 | emp2par->set_connection(par->_rvc(), par->_rv().copy_t(-1)); |
---|
129 | |
---|
130 | dimc = rvc._dsize(); |
---|
131 | }; |
---|
132 | }; |
---|
133 | UIREGISTER(MarginalizedParticle); |
---|
134 | |
---|
135 | //! class used in PF |
---|
136 | class BootstrapParticle : public BM{ |
---|
137 | dirac est; |
---|
138 | shared_ptr<pdf> par; |
---|
139 | shared_ptr<pdf> obs; |
---|
140 | shared_ptr<datalink_part> cond2par; |
---|
141 | shared_ptr<datalink_part> cond2obs; |
---|
142 | shared_ptr<datalink_part> xt2obs; |
---|
143 | shared_ptr<datalink_part> xtm2par; |
---|
144 | public: |
---|
145 | BM* _copy() const{return new BootstrapParticle(*this);}; |
---|
146 | void bayes(const vec &dt, const vec &cond){ |
---|
147 | vec par_cond(par->dimensionc()); |
---|
148 | cond2par->filldown(cond,par_cond); // copy ut |
---|
149 | xtm2par->filldown(est._point(),par_cond); // copy xt-1 |
---|
150 | |
---|
151 | //sample new particle |
---|
152 | est.set_point(par->samplecond(par_cond)); |
---|
153 | //if (evalll) |
---|
154 | vec obs_cond(obs->dimensionc()); |
---|
155 | cond2obs->filldown(cond, obs_cond);// set e.g. ut |
---|
156 | xt2obs->filldown(est._point(), obs_cond);// set e.g. ut |
---|
157 | ll=obs->evallogcond(dt,obs_cond); |
---|
158 | } |
---|
159 | const dirac& posterior() const {return est;} |
---|
160 | |
---|
161 | void set_prior(const epdf *pdf0){est.set_point(pdf0->sample());} |
---|
162 | |
---|
163 | /*! parse structure |
---|
164 | \code |
---|
165 | class = "BootstrapParticle"; |
---|
166 | parameter_pdf = {class = 'epdf_offspring', ...}; |
---|
167 | observation_pdf = {class = 'epdf_offspring',...}; |
---|
168 | \endcode |
---|
169 | If rvs are set, then it checks for compatibility. |
---|
170 | */ |
---|
171 | void from_setting(const Setting &set){ |
---|
172 | BM::from_setting ( set ); |
---|
173 | par = UI::build<pdf> ( set, "parameter_pdf", UI::compulsory ); |
---|
174 | obs = UI::build<pdf> ( set, "observation_pdf", UI::compulsory ); |
---|
175 | } |
---|
176 | void validate(){ |
---|
177 | yrv = obs->_rv(); |
---|
178 | dimy = obs->dimension(); |
---|
179 | set_rv( par->_rv()); |
---|
180 | set_dim( par->dimension()); |
---|
181 | |
---|
182 | rvc = par->_rvc().subt(par->_rv().copy_t(-1)); |
---|
183 | rvc.add(obs->_rvc()); // |
---|
184 | |
---|
185 | cond2obs=new datalink_part; |
---|
186 | cond2par=new datalink_part; |
---|
187 | xt2obs =new datalink_part; |
---|
188 | xtm2par =new datalink_part; |
---|
189 | cond2obs->set_connection(obs->_rvc(), rvc); |
---|
190 | cond2par->set_connection(par->_rvc(), rvc); |
---|
191 | xt2obs->set_connection(obs->_rvc(), _rv()); |
---|
192 | xtm2par->set_connection(par->_rvc(), _rv().copy_t(-1)); |
---|
193 | |
---|
194 | dimc = rvc._dsize(); |
---|
195 | }; |
---|
196 | }; |
---|
197 | UIREGISTER(BootstrapParticle); |
---|
198 | |
---|
199 | |
---|
200 | /*! |
---|
201 | * \brief Trivial particle filter with proposal density equal to parameter evolution model. |
---|
202 | |
---|
203 | Posterior density is represented by a weighted empirical density (\c eEmp ). |
---|
204 | */ |
---|
205 | |
---|
206 | class PF : public BM { |
---|
207 | //! \var log_level_enums weights |
---|
208 | //! all weightes will be logged |
---|
209 | |
---|
210 | //! \var log_level_enums menas |
---|
211 | //! means of particles will be logged |
---|
212 | LOG_LEVEL(PF,logweights,logmeans,logvars); |
---|
213 | |
---|
214 | class pf_mix: public emix_base{ |
---|
215 | Array<BM*> &bms; |
---|
216 | public: |
---|
217 | pf_mix(vec &w0, Array<BM*> &bms0):emix_base(w0),bms(bms0){} |
---|
218 | const epdf* component(const int &i)const{return &(bms(i)->posterior());} |
---|
219 | int no_coms() const {return bms.length();} |
---|
220 | }; |
---|
221 | protected: |
---|
222 | //!number of particles; |
---|
223 | int n; |
---|
224 | //!posterior density |
---|
225 | pf_mix est; |
---|
226 | //! weights; |
---|
227 | vec w; |
---|
228 | //! particles |
---|
229 | Array<BM*> particles; |
---|
230 | //! internal structure storing loglikelihood of predictions |
---|
231 | vec lls; |
---|
232 | |
---|
233 | //! which resampling method will be used |
---|
234 | RESAMPLING_METHOD resmethod; |
---|
235 | //! resampling threshold; in this case its meaning is minimum ratio of active particles |
---|
236 | //! For example, for 0.5 resampling is performed when the numebr of active aprticles drops belo 50%. |
---|
237 | double res_threshold; |
---|
238 | |
---|
239 | //! \name Options |
---|
240 | //!@{ |
---|
241 | //!@} |
---|
242 | |
---|
243 | public: |
---|
244 | //! \name Constructors |
---|
245 | //!@{ |
---|
246 | PF ( ) : est(w,particles) { }; |
---|
247 | |
---|
248 | void set_parameters ( int n0, double res_th0 = 0.5, RESAMPLING_METHOD rm = SYSTEMATIC ) { |
---|
249 | n = n0; |
---|
250 | res_threshold = res_th0; |
---|
251 | resmethod = rm; |
---|
252 | }; |
---|
253 | void set_model ( const BM *particle0, const epdf *prior) { |
---|
254 | if (n>0){ |
---|
255 | particles.set_length(n); |
---|
256 | for (int i=0; i<n;i++){ |
---|
257 | particles(i) = particle0->_copy(); |
---|
258 | particles(i)->set_prior(prior); |
---|
259 | } |
---|
260 | } |
---|
261 | // set values for posterior |
---|
262 | est.set_rv ( particle0->posterior()._rv() ); |
---|
263 | }; |
---|
264 | void set_statistics ( const vec w0, const epdf &epdf0 ) { |
---|
265 | //est.set_statistics ( w0, epdf0 ); |
---|
266 | }; |
---|
267 | /* void set_statistics ( const eEmp &epdf0 ) { |
---|
268 | bdm_assert_debug ( epdf0._rv().equal ( par->_rv() ), "Incompatible input" ); |
---|
269 | est = epdf0; |
---|
270 | };*/ |
---|
271 | //!@} |
---|
272 | |
---|
273 | //! bayes compute weights of the |
---|
274 | virtual void bayes_weights(); |
---|
275 | //! important part of particle filtering - decide if it is time to perform resampling |
---|
276 | virtual bool do_resampling() { |
---|
277 | double eff = 1.0 / ( w * w ); |
---|
278 | return eff < ( res_threshold*n ); |
---|
279 | } |
---|
280 | void bayes ( const vec &yt, const vec &cond ); |
---|
281 | //!access function |
---|
282 | vec& _lls() { |
---|
283 | return lls; |
---|
284 | } |
---|
285 | //!access function |
---|
286 | RESAMPLING_METHOD _resmethod() const { |
---|
287 | return resmethod; |
---|
288 | } |
---|
289 | //! return correctly typed posterior (covariant return) |
---|
290 | const pf_mix& posterior() const { |
---|
291 | return est; |
---|
292 | } |
---|
293 | |
---|
294 | /*! configuration structure for basic PF |
---|
295 | \code |
---|
296 | parameter_pdf = pdf_class; // parameter evolution pdf |
---|
297 | observation_pdf = pdf_class; // observation pdf |
---|
298 | prior = epdf_class; // prior probability density |
---|
299 | --- optional --- |
---|
300 | n = 10; // number of particles |
---|
301 | resmethod = 'systematic', or 'multinomial', or 'stratified' |
---|
302 | // resampling method |
---|
303 | res_threshold = 0.5; // resample when active particles drop below 50% |
---|
304 | \endcode |
---|
305 | */ |
---|
306 | void from_setting ( const Setting &set ) { |
---|
307 | BM::from_setting ( set ); |
---|
308 | UI::get ( log_level, set, "log_level", UI::optional ); |
---|
309 | |
---|
310 | shared_ptr<BM> bm0 = UI::build<BM>(set, "particle",UI::compulsory); |
---|
311 | |
---|
312 | shared_ptr<epdf> pri = UI::build<epdf> ( set, "prior", UI::compulsory ); |
---|
313 | n =0; |
---|
314 | UI::get(n,set,"n",UI::optional);; |
---|
315 | if (n>0){ |
---|
316 | particles.set_length(n); |
---|
317 | for(int i=0;i<n;i++){particles(i)=bm0->_copy();} |
---|
318 | w = ones(n)/n; |
---|
319 | } |
---|
320 | set_prior(pri.get()); |
---|
321 | // set resampling method |
---|
322 | resmethod_from_set ( set ); |
---|
323 | //set drv |
---|
324 | |
---|
325 | rvc = bm0->_rvc(); |
---|
326 | dimc = bm0->dimensionc(); |
---|
327 | BM::set_rv(bm0->_rv()); |
---|
328 | yrv=bm0->_yrv(); |
---|
329 | dimy = bm0->dimensiony(); |
---|
330 | } |
---|
331 | |
---|
332 | void log_register ( bdm::logger& L, const string& prefix ){ |
---|
333 | BM::log_register(L,prefix); |
---|
334 | if (log_level[logweights]){ |
---|
335 | L.add_vector( log_level, logweights, RV ( particles.length()), prefix); |
---|
336 | } |
---|
337 | if (log_level[logmeans]){ |
---|
338 | for (int i=0; i<particles.length(); i++){ |
---|
339 | L.add_vector( log_level, logmeans, RV ( particles(i)->dimension() ), prefix , i); |
---|
340 | } |
---|
341 | } |
---|
342 | if (log_level[logvars]){ |
---|
343 | for (int i=0; i<particles.length(); i++){ |
---|
344 | L.add_vector( log_level, logvars, RV ( particles(i)->dimension() ), prefix , i); |
---|
345 | } |
---|
346 | } |
---|
347 | }; |
---|
348 | void log_write ( ) const { |
---|
349 | BM::log_write(); |
---|
350 | if (log_level[logweights]){ |
---|
351 | log_level.store( logweights, w); |
---|
352 | } |
---|
353 | if (log_level[logmeans]){ |
---|
354 | for (int i=0; i<particles.length(); i++){ |
---|
355 | log_level.store( logmeans, particles(i)->posterior().mean(), i); |
---|
356 | } |
---|
357 | } |
---|
358 | if (log_level[logvars]){ |
---|
359 | for (int i=0; i<particles.length(); i++){ |
---|
360 | log_level.store( logvars, particles(i)->posterior().variance(), i); |
---|
361 | } |
---|
362 | } |
---|
363 | |
---|
364 | } |
---|
365 | |
---|
366 | void set_prior(const epdf *pri){ |
---|
367 | const emix_base *emi=dynamic_cast<const emix_base*>(pri); |
---|
368 | if (emi) { |
---|
369 | bdm_assert(particles.length()>0, "initial particle is not assigned"); |
---|
370 | n = emi->_w().length(); |
---|
371 | int old_n = particles.length(); |
---|
372 | if (n!=old_n){ |
---|
373 | particles.set_length(n,true); |
---|
374 | } |
---|
375 | for(int i=old_n;i<n;i++){particles(i)=particles(0)->_copy();} |
---|
376 | |
---|
377 | for (int i =0; i<n; i++){ |
---|
378 | particles(i)->set_prior(emi->_com(i)); |
---|
379 | } |
---|
380 | } else { |
---|
381 | // try to find "n" |
---|
382 | bdm_assert(n>0, "Field 'n' must be filled when prior is not of type emix"); |
---|
383 | for (int i =0; i<n; i++){ |
---|
384 | particles(i)->set_prior(pri); |
---|
385 | } |
---|
386 | |
---|
387 | } |
---|
388 | } |
---|
389 | //! auxiliary function reading parameter 'resmethod' from configuration file |
---|
390 | void resmethod_from_set ( const Setting &set ) { |
---|
391 | string resmeth; |
---|
392 | if ( UI::get ( resmeth, set, "resmethod", UI::optional ) ) { |
---|
393 | if ( resmeth == "systematic" ) { |
---|
394 | resmethod = SYSTEMATIC; |
---|
395 | } else { |
---|
396 | if ( resmeth == "multinomial" ) { |
---|
397 | resmethod = MULTINOMIAL; |
---|
398 | } else { |
---|
399 | if ( resmeth == "stratified" ) { |
---|
400 | resmethod = STRATIFIED; |
---|
401 | } else { |
---|
402 | bdm_error ( "Unknown resampling method" ); |
---|
403 | } |
---|
404 | } |
---|
405 | } |
---|
406 | } else { |
---|
407 | resmethod = SYSTEMATIC; |
---|
408 | }; |
---|
409 | if ( !UI::get ( res_threshold, set, "res_threshold", UI::optional ) ) { |
---|
410 | res_threshold = 0.9; |
---|
411 | } |
---|
412 | //validate(); |
---|
413 | } |
---|
414 | |
---|
415 | void validate() { |
---|
416 | BM::validate(); |
---|
417 | est.validate(); |
---|
418 | bdm_assert ( n>0, "empty particle pool" ); |
---|
419 | n = w.length(); |
---|
420 | lls = zeros ( n ); |
---|
421 | |
---|
422 | if ( particles(0)->_rv()._dsize() > 0 ) { |
---|
423 | bdm_assert ( particles(0)->_rv()._dsize() == est.dimension(), "Mismatch of RV and dimension of posterior" ); |
---|
424 | } |
---|
425 | } |
---|
426 | //! resample posterior density (from outside - see MPF) |
---|
427 | void resample ( ) { |
---|
428 | ivec ind = zeros_i ( n ); |
---|
429 | bdm::resample(w,ind,resmethod); |
---|
430 | // copy the internals according to ind |
---|
431 | for (int i = 0; i < n; i++ ) { |
---|
432 | if ( ind ( i ) != i ) { |
---|
433 | particles( i ) = particles( ind ( i ) )->_copy(); |
---|
434 | } |
---|
435 | w ( i ) = 1.0 / n; |
---|
436 | } |
---|
437 | } |
---|
438 | //! access function |
---|
439 | Array<BM*>& _particles() { |
---|
440 | return particles; |
---|
441 | } |
---|
442 | |
---|
443 | }; |
---|
444 | UIREGISTER ( PF ); |
---|
445 | |
---|
446 | /*! |
---|
447 | \brief Marginalized Particle filter |
---|
448 | |
---|
449 | A composition of particle filter with exact (or almost exact) bayesian models (BMs). |
---|
450 | The Bayesian models provide marginalized predictive density. Internaly this is achieved by virtual class MPFpdf. |
---|
451 | */ |
---|
452 | |
---|
453 | // class MPF : public BM { |
---|
454 | // //! Introduces new option |
---|
455 | // //! \li means - meaning TODO |
---|
456 | // LOG_LEVEL(MPF,means); |
---|
457 | // protected: |
---|
458 | // //! particle filter on non-linear variable |
---|
459 | // shared_ptr<PF> pf; |
---|
460 | // //! Array of Bayesian models |
---|
461 | // Array<BM*> BMs; |
---|
462 | // |
---|
463 | // //! internal class for pdf providing composition of eEmp with external components |
---|
464 | // |
---|
465 | // class mpfepdf : public epdf { |
---|
466 | // //! pointer to particle filter |
---|
467 | // shared_ptr<PF> &pf; |
---|
468 | // //! pointer to Array of BMs |
---|
469 | // Array<BM*> &BMs; |
---|
470 | // public: |
---|
471 | // //! constructor |
---|
472 | // mpfepdf ( shared_ptr<PF> &pf0, Array<BM*> &BMs0 ) : epdf(), pf ( pf0 ), BMs ( BMs0 ) { }; |
---|
473 | // //! a variant of set parameters - this time, parameters are read from BMs and pf |
---|
474 | // void read_parameters() { |
---|
475 | // rv = concat ( pf->posterior()._rv(), BMs ( 0 )->posterior()._rv() ); |
---|
476 | // dim = pf->posterior().dimension() + BMs ( 0 )->posterior().dimension(); |
---|
477 | // bdm_assert_debug ( dim == rv._dsize(), "Wrong name " ); |
---|
478 | // } |
---|
479 | // vec mean() const; |
---|
480 | // |
---|
481 | // vec variance() const; |
---|
482 | // |
---|
483 | // void qbounds ( vec &lb, vec &ub, double perc = 0.95 ) const; |
---|
484 | // |
---|
485 | // vec sample() const NOT_IMPLEMENTED(0); |
---|
486 | // |
---|
487 | // double evallog ( const vec &val ) const NOT_IMPLEMENTED(0); |
---|
488 | // }; |
---|
489 | // |
---|
490 | // //! Density joining PF.est with conditional parts |
---|
491 | // mpfepdf jest; |
---|
492 | // |
---|
493 | // //! datalink from global yt and cond (Up) to BMs yt and cond (Down) |
---|
494 | // datalink_m2m this2bm; |
---|
495 | // //! datalink from global yt and cond (Up) to PFs yt and cond (Down) |
---|
496 | // datalink_m2m this2pf; |
---|
497 | // //!datalink from PF part to BM |
---|
498 | // datalink_part pf2bm; |
---|
499 | // |
---|
500 | // public: |
---|
501 | // //! Default constructor. |
---|
502 | // MPF () : jest ( pf, BMs ) {}; |
---|
503 | // //! set all parameters at once |
---|
504 | // void set_pf ( shared_ptr<pdf> par0, int n0, RESAMPLING_METHOD rm = SYSTEMATIC ) { |
---|
505 | // if (!pf) pf=new PF; |
---|
506 | // pf->set_model ( par0, par0 ); // <=== nasty!!! |
---|
507 | // pf->set_parameters ( n0, rm ); |
---|
508 | // pf->set_rv(par0->_rv()); |
---|
509 | // BMs.set_length ( n0 ); |
---|
510 | // } |
---|
511 | // //! set a prototype of BM, copy it to as many times as there is particles in pf |
---|
512 | // void set_BM ( const BM &BMcond0 ) { |
---|
513 | // |
---|
514 | // int n = pf->__w().length(); |
---|
515 | // BMs.set_length ( n ); |
---|
516 | // // copy |
---|
517 | // //BMcond0 .condition ( pf->posterior()._sample ( 0 ) ); |
---|
518 | // for ( int i = 0; i < n; i++ ) { |
---|
519 | // BMs ( i ) = (BM*) BMcond0._copy(); |
---|
520 | // } |
---|
521 | // }; |
---|
522 | // |
---|
523 | // void bayes ( const vec &yt, const vec &cond ); |
---|
524 | // |
---|
525 | // const epdf& posterior() const { |
---|
526 | // return jest; |
---|
527 | // } |
---|
528 | // |
---|
529 | // //!Access function |
---|
530 | // const BM* _BM ( int i ) { |
---|
531 | // return BMs ( i ); |
---|
532 | // } |
---|
533 | // PF& _pf() {return *pf;} |
---|
534 | // |
---|
535 | // |
---|
536 | // virtual double logpred ( const vec &yt ) const NOT_IMPLEMENTED(0); |
---|
537 | // |
---|
538 | // virtual epdf* epredictor() const NOT_IMPLEMENTED(NULL); |
---|
539 | // |
---|
540 | // virtual pdf* predictor() const NOT_IMPLEMENTED(NULL); |
---|
541 | // |
---|
542 | // |
---|
543 | // /*! configuration structure for basic PF |
---|
544 | // \code |
---|
545 | // BM = BM_class; // Bayesian filtr for analytical part of the model |
---|
546 | // parameter_pdf = pdf_class; // transitional pdf for non-parametric part of the model |
---|
547 | // prior = epdf_class; // prior probability density |
---|
548 | // --- optional --- |
---|
549 | // n = 10; // number of particles |
---|
550 | // resmethod = 'systematic', or 'multinomial', or 'stratified' |
---|
551 | // // resampling method |
---|
552 | // \endcode |
---|
553 | // */ |
---|
554 | // void from_setting ( const Setting &set ) { |
---|
555 | // BM::from_setting( set ); |
---|
556 | // |
---|
557 | // shared_ptr<pdf> par = UI::build<pdf> ( set, "parameter_pdf", UI::compulsory ); |
---|
558 | // |
---|
559 | // pf = new PF; |
---|
560 | // // prior must be set before BM |
---|
561 | // pf->prior_from_set ( set ); |
---|
562 | // pf->resmethod_from_set ( set ); |
---|
563 | // pf->set_model ( par, par ); // too hackish! |
---|
564 | // |
---|
565 | // shared_ptr<BM> BM0 = UI::build<BM> ( set, "BM", UI::compulsory ); |
---|
566 | // set_BM ( *BM0 ); |
---|
567 | // |
---|
568 | // //set drv |
---|
569 | // //??set_yrv(concat(BM0->_yrv(),u) ); |
---|
570 | // set_yrv ( BM0->_yrv() ); |
---|
571 | // rvc = BM0->_rvc().subt ( par->_rv() ); |
---|
572 | // //find potential input - what remains in rvc when we subtract rv |
---|
573 | // RV u = par->_rvc().subt ( par->_rv().copy_t ( -1 ) ); |
---|
574 | // rvc.add ( u ); |
---|
575 | // dimc = rvc._dsize(); |
---|
576 | // validate(); |
---|
577 | // } |
---|
578 | // |
---|
579 | // void validate() { |
---|
580 | // BM::validate(); |
---|
581 | // try { |
---|
582 | // pf->validate(); |
---|
583 | // } catch ( std::exception ) { |
---|
584 | // throw UIException ( "Error in PF part of MPF:" ); |
---|
585 | // } |
---|
586 | // jest.read_parameters(); |
---|
587 | // this2bm.set_connection ( BMs ( 0 )->_yrv(), BMs ( 0 )->_rvc(), yrv, rvc ); |
---|
588 | // this2pf.set_connection ( pf->_yrv(), pf->_rvc(), yrv, rvc ); |
---|
589 | // pf2bm.set_connection ( BMs ( 0 )->_rvc(), pf->posterior()._rv() ); |
---|
590 | // } |
---|
591 | // }; |
---|
592 | // UIREGISTER ( MPF ); |
---|
593 | |
---|
594 | /*! ARXg for estimation of state-space variances |
---|
595 | */ |
---|
596 | // class MPF_ARXg :public BM{ |
---|
597 | // protected: |
---|
598 | // shared_ptr<PF> pf; |
---|
599 | // //! pointer to Array of BMs |
---|
600 | // Array<ARX*> BMso; |
---|
601 | // //! pointer to Array of BMs |
---|
602 | // Array<ARX*> BMsp; |
---|
603 | // //!parameter evolution |
---|
604 | // shared_ptr<fnc> g; |
---|
605 | // //!observation function |
---|
606 | // shared_ptr<fnc> h; |
---|
607 | // |
---|
608 | // public: |
---|
609 | // void bayes(const vec &yt, const vec &cond ); |
---|
610 | // void from_setting(const Setting &set) ; |
---|
611 | // void validate() { |
---|
612 | // bdm_assert(g->dimensionc()==g->dimension(),"not supported yet"); |
---|
613 | // bdm_assert(h->dimensionc()==g->dimension(),"not supported yet"); |
---|
614 | // } |
---|
615 | // |
---|
616 | // double logpred(const vec &cond) const NOT_IMPLEMENTED(0.0); |
---|
617 | // epdf* epredictor() const NOT_IMPLEMENTED(NULL); |
---|
618 | // pdf* predictor() const NOT_IMPLEMENTED(NULL); |
---|
619 | // const epdf& posterior() const {return pf->posterior();}; |
---|
620 | // |
---|
621 | // void log_register( logger &L, const string &prefix ){ |
---|
622 | // BM::log_register(L,prefix); |
---|
623 | // registered_logger->ids.set_size ( 3 ); |
---|
624 | // registered_logger->ids(1)= L.add_vector(RV("Q",dimension()*dimension()), prefix+L.prefix_sep()+"Q"); |
---|
625 | // registered_logger->ids(2)= L.add_vector(RV("R",dimensiony()*dimensiony()), prefix+L.prefix_sep()+"R"); |
---|
626 | // |
---|
627 | // }; |
---|
628 | // void log_write() const { |
---|
629 | // BM::log_write(); |
---|
630 | // mat mQ=zeros(dimension(),dimension()); |
---|
631 | // mat pom=zeros(dimension(),dimension()); |
---|
632 | // mat mR=zeros(dimensiony(),dimensiony()); |
---|
633 | // mat pom2=zeros(dimensiony(),dimensiony()); |
---|
634 | // mat dum; |
---|
635 | // const vec w=pf->posterior()._w(); |
---|
636 | // for (int i=0; i<w.length(); i++){ |
---|
637 | // BMsp(i)->posterior().mean_mat(dum,pom); |
---|
638 | // mQ += w(i) * pom; |
---|
639 | // BMso(i)->posterior().mean_mat(dum,pom2); |
---|
640 | // mR += w(i) * pom2; |
---|
641 | // |
---|
642 | // } |
---|
643 | // registered_logger->L.log_vector ( registered_logger->ids ( 1 ), cvectorize(mQ) ); |
---|
644 | // registered_logger->L.log_vector ( registered_logger->ids ( 2 ), cvectorize(mR) ); |
---|
645 | // |
---|
646 | // } |
---|
647 | // }; |
---|
648 | // UIREGISTER(MPF_ARXg); |
---|
649 | |
---|
650 | |
---|
651 | } |
---|
652 | #endif // KF_H |
---|
653 | |
---|