1 | /*! |
---|
2 | \file |
---|
3 | \brief Probability distributions for Exponential Family models. |
---|
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 EF_H |
---|
14 | #define EF_H |
---|
15 | |
---|
16 | #include <itpp/itbase.h> |
---|
17 | #include "../math/libDC.h" |
---|
18 | #include "libBM.h" |
---|
19 | #include "../itpp_ext.h" |
---|
20 | //#include <std> |
---|
21 | |
---|
22 | using namespace itpp; |
---|
23 | |
---|
24 | |
---|
25 | //! Global Uniform_RNG |
---|
26 | extern Uniform_RNG UniRNG; |
---|
27 | //! Global Normal_RNG |
---|
28 | extern Normal_RNG NorRNG; |
---|
29 | //! Global Gamma_RNG |
---|
30 | extern Gamma_RNG GamRNG; |
---|
31 | |
---|
32 | /*! |
---|
33 | * \brief General conjugate exponential family posterior density. |
---|
34 | |
---|
35 | * More?... |
---|
36 | */ |
---|
37 | |
---|
38 | class eEF : public epdf { |
---|
39 | public: |
---|
40 | // eEF() :epdf() {}; |
---|
41 | //! default constructor |
---|
42 | eEF ( const RV &rv ) :epdf ( rv ) {}; |
---|
43 | //! logarithm of the normalizing constant, \f$\mathcal{I}\f$ |
---|
44 | virtual double lognc() const =0; |
---|
45 | //!TODO decide if it is really needed |
---|
46 | virtual void dupdate ( mat &v ) {it_error ( "Not implemented" );}; |
---|
47 | //!Evaluate normalized log-probability |
---|
48 | virtual double evalpdflog_nn ( const vec &val ) const{it_error ( "Not implemented" );return 0.0;}; |
---|
49 | //!Evaluate normalized log-probability |
---|
50 | virtual double evalpdflog ( const vec &val ) const {double tmp;tmp= evalpdflog_nn ( val )-lognc();it_assert_debug(std::isfinite(tmp),"why?"); return tmp;} |
---|
51 | //!Evaluate normalized log-probability for many samples |
---|
52 | virtual vec evalpdflog ( const mat &Val ) const { |
---|
53 | vec x ( Val.cols() ); |
---|
54 | for ( int i=0;i<Val.cols();i++ ) {x ( i ) =evalpdflog_nn ( Val.get_col ( i ) ) ;} |
---|
55 | return x-lognc(); |
---|
56 | } |
---|
57 | //!Power of the density, used e.g. to flatten the density |
---|
58 | virtual void pow ( double p ) {it_error ( "Not implemented" );}; |
---|
59 | }; |
---|
60 | |
---|
61 | /*! |
---|
62 | * \brief Exponential family model. |
---|
63 | |
---|
64 | * More?... |
---|
65 | */ |
---|
66 | |
---|
67 | class mEF : public mpdf { |
---|
68 | |
---|
69 | public: |
---|
70 | //! Default constructor |
---|
71 | mEF ( const RV &rv0, const RV &rvc0 ) :mpdf ( rv0,rvc0 ) {}; |
---|
72 | }; |
---|
73 | |
---|
74 | //! Estimator for Exponential family |
---|
75 | class BMEF : public BM { |
---|
76 | protected: |
---|
77 | //! forgetting factor |
---|
78 | double frg; |
---|
79 | //! cached value of lognc() in the previous step (used in evaluation of \c ll ) |
---|
80 | double last_lognc; |
---|
81 | public: |
---|
82 | //! Default constructor |
---|
83 | BMEF ( const RV &rv, double frg0=1.0 ) :BM ( rv ), frg ( frg0 ) {} |
---|
84 | //! Copy constructor |
---|
85 | BMEF ( const BMEF &B ) :BM ( B ), frg ( B.frg ), last_lognc ( B.last_lognc ) {} |
---|
86 | //!get statistics from another model |
---|
87 | virtual void set_statistics ( const BMEF* BM0 ) {it_error ( "Not implemented" );}; |
---|
88 | //! Weighted update of sufficient statistics (Bayes rule) |
---|
89 | virtual void bayes ( const vec &data, const double w ) {}; |
---|
90 | //original Bayes |
---|
91 | void bayes ( const vec &dt ); |
---|
92 | //!Flatten the posterior according to the given BMEF (of the same type!) |
---|
93 | virtual void flatten ( const BMEF * B ) {it_error ( "Not implemented" );} |
---|
94 | //!Flatten the posterior as if to keep nu0 data |
---|
95 | // virtual void flatten ( double nu0 ) {it_error ( "Not implemented" );} |
---|
96 | |
---|
97 | BMEF* _copy_ ( bool changerv=false ) {it_error ( "function _copy_ not implemented for this BM" ); return NULL;}; |
---|
98 | }; |
---|
99 | |
---|
100 | template<class sq_T> |
---|
101 | class mlnorm; |
---|
102 | |
---|
103 | /*! |
---|
104 | * \brief Gaussian density with positive definite (decomposed) covariance matrix. |
---|
105 | |
---|
106 | * More?... |
---|
107 | */ |
---|
108 | template<class sq_T> |
---|
109 | class enorm : public eEF { |
---|
110 | protected: |
---|
111 | //! mean value |
---|
112 | vec mu; |
---|
113 | //! Covariance matrix in decomposed form |
---|
114 | sq_T R; |
---|
115 | //! dimension (redundant from rv.count() for easier coding ) |
---|
116 | int dim; |
---|
117 | public: |
---|
118 | //!Default constructor |
---|
119 | enorm ( const RV &rv ); |
---|
120 | //! Set mean value \c mu and covariance \c R |
---|
121 | void set_parameters ( const vec &mu,const sq_T &R ); |
---|
122 | ////! tupdate in exponential form (not really handy) |
---|
123 | //void tupdate ( double phi, mat &vbar, double nubar ); |
---|
124 | //! dupdate in exponential form (not really handy) |
---|
125 | void dupdate ( mat &v,double nu=1.0 ); |
---|
126 | |
---|
127 | vec sample() const; |
---|
128 | //! TODO is it used? |
---|
129 | mat sample ( int N ) const; |
---|
130 | double eval ( const vec &val ) const ; |
---|
131 | double evalpdflog_nn ( const vec &val ) const; |
---|
132 | double lognc () const; |
---|
133 | vec mean() const {return mu;} |
---|
134 | // mlnorm<sq_T>* condition ( const RV &rvn ) const ; |
---|
135 | mpdf* condition ( const RV &rvn ) const ; |
---|
136 | // enorm<sq_T>* marginal ( const RV &rv ) const; |
---|
137 | epdf* marginal ( const RV &rv ) const; |
---|
138 | //Access methods |
---|
139 | //! returns a pointer to the internal mean value. Use with Care! |
---|
140 | vec& _mu() {return mu;} |
---|
141 | |
---|
142 | //! access function |
---|
143 | void set_mu ( const vec mu0 ) { mu=mu0;} |
---|
144 | |
---|
145 | //! returns pointers to the internal variance and its inverse. Use with Care! |
---|
146 | sq_T& _R() {return R;} |
---|
147 | |
---|
148 | //! access method |
---|
149 | // mat getR () {return R.to_mat();} |
---|
150 | }; |
---|
151 | |
---|
152 | /*! |
---|
153 | * \brief Gauss-inverse-Wishart density stored in LD form |
---|
154 | |
---|
155 | * For \f$p\f$-variate densities, given rv.count() should be \f$p\times\f$ V.rows(). |
---|
156 | * |
---|
157 | */ |
---|
158 | class egiw : public eEF { |
---|
159 | protected: |
---|
160 | //! Extended information matrix of sufficient statistics |
---|
161 | ldmat V; |
---|
162 | //! Number of data records (degrees of freedom) of sufficient statistics |
---|
163 | double nu; |
---|
164 | //! Dimension of the output |
---|
165 | int xdim; |
---|
166 | //! Dimension of the regressor |
---|
167 | int nPsi; |
---|
168 | public: |
---|
169 | //!Default constructor, if nu0<0 a minimal nu0 will be computed |
---|
170 | egiw ( RV rv, mat V0, double nu0=-1.0 ) : eEF ( rv ), V ( V0 ), nu ( nu0 ) { |
---|
171 | xdim = rv.count() /V.rows(); |
---|
172 | it_assert_debug ( rv.count() ==xdim*V.rows(),"Incompatible V0." ); |
---|
173 | nPsi = V.rows()-xdim; |
---|
174 | //set mu to have proper normalization and |
---|
175 | if (nu0<0){ |
---|
176 | nu = 0.1 +nPsi +2*xdim +2; // +2 assures finite expected value of R |
---|
177 | // terms before that are sufficient for finite normalization |
---|
178 | } |
---|
179 | } |
---|
180 | //!Full constructor for V in ldmat form |
---|
181 | egiw ( RV rv, ldmat V0, double nu0=-1.0 ) : eEF ( rv ), V ( V0 ), nu ( nu0 ) { |
---|
182 | xdim = rv.count() /V.rows(); |
---|
183 | it_assert_debug ( rv.count() ==xdim*V.rows(),"Incompatible V0." ); |
---|
184 | nPsi = V.rows()-xdim; |
---|
185 | if (nu0<0){ |
---|
186 | nu = 0.1 +nPsi +2*xdim +2; // +2 assures finite expected value of R |
---|
187 | // terms before that are sufficient for finite normalization |
---|
188 | } |
---|
189 | } |
---|
190 | |
---|
191 | vec sample() const; |
---|
192 | vec mean() const; |
---|
193 | void mean_mat ( mat &M, mat&R ) const; |
---|
194 | //! In this instance, val= [theta, r]. For multivariate instances, it is stored columnwise val = [theta_1 theta_2 ... r_1 r_2 ] |
---|
195 | double evalpdflog_nn ( const vec &val ) const; |
---|
196 | double lognc () const; |
---|
197 | |
---|
198 | //Access |
---|
199 | //! returns a pointer to the internal statistics. Use with Care! |
---|
200 | ldmat& _V() {return V;} |
---|
201 | //! returns a pointer to the internal statistics. Use with Care! |
---|
202 | double& _nu() {return nu;} |
---|
203 | void pow ( double p ) {V*=p;nu*=p;}; |
---|
204 | }; |
---|
205 | |
---|
206 | /*! \brief Dirichlet posterior density |
---|
207 | |
---|
208 | Continuous Dirichlet density of \f$n\f$-dimensional variable \f$x\f$ |
---|
209 | \f[ |
---|
210 | f(x|\beta) = \frac{\Gamma[\gamma]}{\prod_{i=1}^{n}\Gamma(\beta_i)} \prod_{i=1}^{n}x_i^{\beta_i-1} |
---|
211 | \f] |
---|
212 | where \f$\gamma=\sum_i \beta_i\f$. |
---|
213 | */ |
---|
214 | class eDirich: public eEF { |
---|
215 | protected: |
---|
216 | //!sufficient statistics |
---|
217 | vec beta; |
---|
218 | public: |
---|
219 | //!Default constructor |
---|
220 | eDirich ( const RV &rv, const vec &beta0 ) : eEF ( rv ),beta ( beta0 ) {it_assert_debug ( rv.count() ==beta.length(),"Incompatible statistics" ); }; |
---|
221 | //! Copy constructor |
---|
222 | eDirich ( const eDirich &D0 ) : eEF ( D0.rv ),beta ( D0.beta ) {}; |
---|
223 | vec sample() const {it_error ( "Not implemented" );return vec_1 ( 0.0 );}; |
---|
224 | vec mean() const {return beta/sum ( beta );}; |
---|
225 | //! In this instance, val is ... |
---|
226 | double evalpdflog_nn ( const vec &val ) const {return ( beta-1 ) *log ( val );}; |
---|
227 | double lognc () const { |
---|
228 | double gam=sum ( beta ); |
---|
229 | double lgb=0.0; |
---|
230 | for ( int i=0;i<beta.length();i++ ) {lgb+=lgamma ( beta ( i ) );} |
---|
231 | return lgb-lgamma ( gam ); |
---|
232 | }; |
---|
233 | //!access function |
---|
234 | vec& _beta() {return beta;} |
---|
235 | //!Set internal parameters |
---|
236 | void set_parameters ( const vec &beta0 ) { |
---|
237 | if ( beta0.length() !=beta.length() ) { |
---|
238 | it_assert_debug ( rv.length() ==1,"Undefined" ); |
---|
239 | rv.set_size ( 0,beta0.length() ); |
---|
240 | } |
---|
241 | beta= beta0; |
---|
242 | } |
---|
243 | }; |
---|
244 | |
---|
245 | //! \brief Estimator for Multinomial density |
---|
246 | class multiBM : public BMEF { |
---|
247 | protected: |
---|
248 | //! Conjugate prior and posterior |
---|
249 | eDirich est; |
---|
250 | //! Pointer inside est to sufficient statistics |
---|
251 | vec β |
---|
252 | public: |
---|
253 | //!Default constructor |
---|
254 | multiBM ( const RV &rv, const vec beta0 ) : BMEF ( rv ),est ( rv,beta0 ),beta ( est._beta() ) {last_lognc=est.lognc();} |
---|
255 | //!Copy constructor |
---|
256 | multiBM ( const multiBM &B ) : BMEF ( B ),est ( rv,B.beta ),beta ( est._beta() ) {} |
---|
257 | //! Sets sufficient statistics to match that of givefrom mB0 |
---|
258 | void set_statistics ( const BM* mB0 ) {const multiBM* mB=dynamic_cast<const multiBM*> ( mB0 ); beta=mB->beta;} |
---|
259 | void bayes ( const vec &dt ) { |
---|
260 | if ( frg<1.0 ) {beta*=frg;last_lognc=est.lognc();} |
---|
261 | beta+=dt; |
---|
262 | if ( evalll ) {ll=est.lognc()-last_lognc;} |
---|
263 | } |
---|
264 | double logpred ( const vec &dt ) const { |
---|
265 | eDirich pred ( est ); |
---|
266 | vec &beta = pred._beta(); |
---|
267 | |
---|
268 | double lll; |
---|
269 | if ( frg<1.0 ) |
---|
270 | {beta*=frg;lll=pred.lognc();} |
---|
271 | else |
---|
272 | if ( evalll ) {lll=last_lognc;} |
---|
273 | else{lll=pred.lognc();} |
---|
274 | |
---|
275 | beta+=dt; |
---|
276 | return pred.lognc()-lll; |
---|
277 | } |
---|
278 | void flatten ( const BMEF* B ) { |
---|
279 | const multiBM* E=dynamic_cast<const multiBM*> ( B ); |
---|
280 | // sum(beta) should be equal to sum(B.beta) |
---|
281 | const vec &Eb=E->beta;//const_cast<multiBM*> ( E )->_beta(); |
---|
282 | beta*= ( sum ( Eb ) /sum ( beta ) ); |
---|
283 | if ( evalll ) {last_lognc=est.lognc();} |
---|
284 | } |
---|
285 | const epdf& _epdf() const {return est;}; |
---|
286 | const eDirich* _e() const {return &est;}; |
---|
287 | void set_parameters ( const vec &beta0 ) { |
---|
288 | est.set_parameters ( beta0 ); |
---|
289 | rv = est._rv(); |
---|
290 | if ( evalll ) {last_lognc=est.lognc();} |
---|
291 | } |
---|
292 | }; |
---|
293 | |
---|
294 | /*! |
---|
295 | \brief Gamma posterior density |
---|
296 | |
---|
297 | Multivariate Gamma density as product of independent univariate densities. |
---|
298 | \f[ |
---|
299 | f(x|\alpha,\beta) = \prod f(x_i|\alpha_i,\beta_i) |
---|
300 | \f] |
---|
301 | */ |
---|
302 | |
---|
303 | class egamma : public eEF { |
---|
304 | protected: |
---|
305 | //! Vector \f$\alpha\f$ |
---|
306 | vec alpha; |
---|
307 | //! Vector \f$\beta\f$ |
---|
308 | vec beta; |
---|
309 | public : |
---|
310 | //! Default constructor |
---|
311 | egamma ( const RV &rv ) :eEF ( rv ) {}; |
---|
312 | //! Sets parameters |
---|
313 | void set_parameters ( const vec &a, const vec &b ) {alpha=a,beta=b;}; |
---|
314 | vec sample() const; |
---|
315 | //! TODO: is it used anywhere? |
---|
316 | // mat sample ( int N ) const; |
---|
317 | double evalpdflog ( const vec &val ) const; |
---|
318 | double lognc () const; |
---|
319 | //! Returns poiter to alpha and beta. Potentially dengerous: use with care! |
---|
320 | void _param ( vec* &a, vec* &b ) {a=αb=β}; |
---|
321 | vec mean() const {vec pom ( alpha ); pom/=beta; return pom;} |
---|
322 | }; |
---|
323 | /* |
---|
324 | //! Weighted mixture of epdfs with external owned components. |
---|
325 | class emix : public epdf { |
---|
326 | protected: |
---|
327 | int n; |
---|
328 | vec &w; |
---|
329 | Array<epdf*> Coms; |
---|
330 | public: |
---|
331 | //! Default constructor |
---|
332 | emix ( const RV &rv, vec &w0): epdf(rv), n(w0.length()), w(w0), Coms(n) {}; |
---|
333 | void set_parameters( int &i, double wi, epdf* ep){w(i)=wi;Coms(i)=ep;} |
---|
334 | vec mean(){vec pom; for(int i=0;i<n;i++){pom+=Coms(i)->mean()*w(i);} return pom;}; |
---|
335 | vec sample() {it_error ( "Not implemented" );return 0;} |
---|
336 | }; |
---|
337 | */ |
---|
338 | |
---|
339 | //! Uniform distributed density on a rectangular support |
---|
340 | |
---|
341 | class euni: public epdf { |
---|
342 | protected: |
---|
343 | //! lower bound on support |
---|
344 | vec low; |
---|
345 | //! upper bound on support |
---|
346 | vec high; |
---|
347 | //! internal |
---|
348 | vec distance; |
---|
349 | //! normalizing coefficients |
---|
350 | double nk; |
---|
351 | //! cache of log( \c nk ) |
---|
352 | double lnk; |
---|
353 | public: |
---|
354 | //! Defualt constructor |
---|
355 | euni ( const RV rv ) :epdf ( rv ) {} |
---|
356 | double eval ( const vec &val ) const {return nk;} |
---|
357 | double evalpdflog ( const vec &val ) const {return lnk;} |
---|
358 | vec sample() const { |
---|
359 | vec smp ( rv.count() ); |
---|
360 | #pragma omp critical |
---|
361 | UniRNG.sample_vector ( rv.count(),smp ); |
---|
362 | return low+elem_mult ( distance,smp ); |
---|
363 | } |
---|
364 | //! set values of \c low and \c high |
---|
365 | void set_parameters ( const vec &low0, const vec &high0 ) { |
---|
366 | distance = high0-low0; |
---|
367 | it_assert_debug ( min ( distance ) >0.0,"bad support" ); |
---|
368 | low = low0; |
---|
369 | high = high0; |
---|
370 | nk = prod ( 1.0/distance ); |
---|
371 | lnk = log ( nk ); |
---|
372 | } |
---|
373 | vec mean() const {vec pom=high; pom-=low; pom/=2.0; return pom;} |
---|
374 | }; |
---|
375 | |
---|
376 | |
---|
377 | /*! |
---|
378 | \brief Normal distributed linear function with linear function of mean value; |
---|
379 | |
---|
380 | Mean value \f$mu=A*rvc+mu_0\f$. |
---|
381 | */ |
---|
382 | template<class sq_T> |
---|
383 | class mlnorm : public mEF { |
---|
384 | protected: |
---|
385 | //! Internal epdf that arise by conditioning on \c rvc |
---|
386 | enorm<sq_T> epdf; |
---|
387 | mat A; |
---|
388 | vec mu_const; |
---|
389 | vec& _mu; //cached epdf.mu; |
---|
390 | public: |
---|
391 | //! Constructor |
---|
392 | mlnorm ( const RV &rv, const RV &rvc ); |
---|
393 | //! Set \c A and \c R |
---|
394 | void set_parameters ( const mat &A, const vec &mu0, const sq_T &R ); |
---|
395 | // //!Generate one sample of the posterior |
---|
396 | // vec samplecond (const vec &cond, double &lik ); |
---|
397 | // //!Generate matrix of samples of the posterior |
---|
398 | // mat samplecond (const vec &cond, vec &lik, int n ); |
---|
399 | //! Set value of \c rvc . Result of this operation is stored in \c epdf use function \c _ep to access it. |
---|
400 | void condition ( const vec &cond ); |
---|
401 | |
---|
402 | //!access function |
---|
403 | vec& _mu_const() {return mu_const;} |
---|
404 | //!access function |
---|
405 | mat& _A() {return A;} |
---|
406 | //!access function |
---|
407 | mat _R() {return epdf._R().to_mat();} |
---|
408 | |
---|
409 | template<class sq_M> |
---|
410 | friend std::ostream &operator<< ( std::ostream &os, mlnorm<sq_M> &ml ); |
---|
411 | }; |
---|
412 | |
---|
413 | /*! (Approximate) Student t density with linear function of mean value |
---|
414 | */ |
---|
415 | class mlstudent : public mlnorm<ldmat> { |
---|
416 | protected: |
---|
417 | ldmat Lambda; |
---|
418 | ldmat &_R; |
---|
419 | ldmat Re; |
---|
420 | public: |
---|
421 | mlstudent ( const RV &rv0, const RV &rvc0 ) :mlnorm<ldmat> ( rv0,rvc0 ), |
---|
422 | Lambda ( rv0.count() ), |
---|
423 | _R ( epdf._R() ) {} |
---|
424 | void set_parameters ( const mat &A0, const vec &mu0, const ldmat &R0, const ldmat& Lambda0) { |
---|
425 | epdf.set_parameters ( zeros ( rv.count() ),Lambda ); |
---|
426 | A = A0; |
---|
427 | mu_const = mu0; |
---|
428 | Re=R0; |
---|
429 | Lambda = Lambda0; |
---|
430 | } |
---|
431 | void condition ( const vec &cond ) { |
---|
432 | _mu = A*cond + mu_const; |
---|
433 | double zeta; |
---|
434 | //ugly hack! |
---|
435 | if ((cond.length()+1)==Lambda.rows()){ |
---|
436 | zeta = Lambda.invqform ( concat(cond, vec_1(1.0)) ); |
---|
437 | } else { |
---|
438 | zeta = Lambda.invqform ( cond ); |
---|
439 | } |
---|
440 | _R = Re; |
---|
441 | _R*=( 1+zeta );// / ( nu ); << nu is in Re!!!!!! |
---|
442 | }; |
---|
443 | |
---|
444 | }; |
---|
445 | /*! |
---|
446 | \brief Gamma random walk |
---|
447 | |
---|
448 | Mean value, \f$\mu\f$, of this density is given by \c rvc . |
---|
449 | Standard deviation of the random walk is proportional to one \f$k\f$-th the mean. |
---|
450 | This is achieved by setting \f$\alpha=k\f$ and \f$\beta=k/\mu\f$. |
---|
451 | |
---|
452 | The standard deviation of the walk is then: \f$\mu/\sqrt(k)\f$. |
---|
453 | */ |
---|
454 | class mgamma : public mEF { |
---|
455 | protected: |
---|
456 | //! Internal epdf that arise by conditioning on \c rvc |
---|
457 | egamma epdf; |
---|
458 | //! Constant \f$k\f$ |
---|
459 | double k; |
---|
460 | //! cache of epdf.beta |
---|
461 | vec* _beta; |
---|
462 | |
---|
463 | public: |
---|
464 | //! Constructor |
---|
465 | mgamma ( const RV &rv,const RV &rvc ); |
---|
466 | //! Set value of \c k |
---|
467 | void set_parameters ( double k ); |
---|
468 | void condition ( const vec &val ) {*_beta=k/val;}; |
---|
469 | }; |
---|
470 | |
---|
471 | /*! |
---|
472 | \brief Gamma random walk around a fixed point |
---|
473 | |
---|
474 | Mean value, \f$\mu\f$, of this density is given by a geometric combination of \c rvc and given fixed point, \f$p\f$. \f$l\f$ is the coefficient of the geometric combimation |
---|
475 | \f[ \mu = \mu_{t-1} ^{l} p^{1-l}\f] |
---|
476 | |
---|
477 | Standard deviation of the random walk is proportional to one \f$k\f$-th the mean. |
---|
478 | This is achieved by setting \f$\alpha=k\f$ and \f$\beta=k/\mu\f$. |
---|
479 | |
---|
480 | The standard deviation of the walk is then: \f$\mu/\sqrt(k)\f$. |
---|
481 | */ |
---|
482 | class mgamma_fix : public mgamma { |
---|
483 | protected: |
---|
484 | //! parameter l |
---|
485 | double l; |
---|
486 | //! reference vector |
---|
487 | vec refl; |
---|
488 | public: |
---|
489 | //! Constructor |
---|
490 | mgamma_fix ( const RV &rv,const RV &rvc ) : mgamma ( rv,rvc ),refl ( rv.count() ) {}; |
---|
491 | //! Set value of \c k |
---|
492 | void set_parameters ( double k0 , vec ref0, double l0 ) { |
---|
493 | mgamma::set_parameters ( k0 ); |
---|
494 | refl=pow ( ref0,1.0-l0 );l=l0; |
---|
495 | }; |
---|
496 | |
---|
497 | void condition ( const vec &val ) {vec mean=elem_mult ( refl,pow ( val,l ) ); *_beta=k/mean;}; |
---|
498 | }; |
---|
499 | |
---|
500 | //! Switch between various resampling methods. |
---|
501 | enum RESAMPLING_METHOD { MULTINOMIAL = 0, STRATIFIED = 1, SYSTEMATIC = 3 }; |
---|
502 | /*! |
---|
503 | \brief Weighted empirical density |
---|
504 | |
---|
505 | Used e.g. in particle filters. |
---|
506 | */ |
---|
507 | class eEmp: public epdf { |
---|
508 | protected : |
---|
509 | //! Number of particles |
---|
510 | int n; |
---|
511 | //! Sample weights \f$w\f$ |
---|
512 | vec w; |
---|
513 | //! Samples \f$x^{(i)}, i=1..n\f$ |
---|
514 | Array<vec> samples; |
---|
515 | public: |
---|
516 | //! Default constructor |
---|
517 | eEmp ( const RV &rv0 ,int n0 ) :epdf ( rv0 ),n ( n0 ),w ( n ),samples ( n ) {}; |
---|
518 | //! Set samples and weights |
---|
519 | void set_parameters ( const vec &w0, const epdf* pdf0 ); |
---|
520 | //! Set sample |
---|
521 | void set_samples ( const epdf* pdf0 ); |
---|
522 | //! Potentially dangerous, use with care. |
---|
523 | vec& _w() {return w;}; |
---|
524 | //! access function |
---|
525 | Array<vec>& _samples() {return samples;}; |
---|
526 | //! Function performs resampling, i.e. removal of low-weight samples and duplication of high-weight samples such that the new samples represent the same density. |
---|
527 | ivec resample ( RESAMPLING_METHOD method = SYSTEMATIC ); |
---|
528 | //! inherited operation : NOT implemneted |
---|
529 | vec sample() const {it_error ( "Not implemented" );return 0;} |
---|
530 | //! inherited operation : NOT implemneted |
---|
531 | double evalpdflog ( const vec &val ) const {it_error ( "Not implemented" );return 0.0;} |
---|
532 | vec mean() const { |
---|
533 | vec pom=zeros ( rv.count() ); |
---|
534 | for ( int i=0;i<n;i++ ) {pom+=samples ( i ) *w ( i );} |
---|
535 | return pom; |
---|
536 | } |
---|
537 | }; |
---|
538 | |
---|
539 | |
---|
540 | //////////////////////// |
---|
541 | |
---|
542 | template<class sq_T> |
---|
543 | enorm<sq_T>::enorm ( const RV &rv ) :eEF ( rv ), mu ( rv.count() ),R ( rv.count() ),dim ( rv.count() ) {}; |
---|
544 | |
---|
545 | template<class sq_T> |
---|
546 | void enorm<sq_T>::set_parameters ( const vec &mu0, const sq_T &R0 ) { |
---|
547 | //Fixme test dimensions of mu0 and R0; |
---|
548 | mu = mu0; |
---|
549 | R = R0; |
---|
550 | }; |
---|
551 | |
---|
552 | template<class sq_T> |
---|
553 | void enorm<sq_T>::dupdate ( mat &v, double nu ) { |
---|
554 | // |
---|
555 | }; |
---|
556 | |
---|
557 | // template<class sq_T> |
---|
558 | // void enorm<sq_T>::tupdate ( double phi, mat &vbar, double nubar ) { |
---|
559 | // // |
---|
560 | // }; |
---|
561 | |
---|
562 | template<class sq_T> |
---|
563 | vec enorm<sq_T>::sample() const { |
---|
564 | vec x ( dim ); |
---|
565 | NorRNG.sample_vector ( dim,x ); |
---|
566 | vec smp = R.sqrt_mult ( x ); |
---|
567 | |
---|
568 | smp += mu; |
---|
569 | return smp; |
---|
570 | }; |
---|
571 | |
---|
572 | template<class sq_T> |
---|
573 | mat enorm<sq_T>::sample ( int N ) const { |
---|
574 | mat X ( dim,N ); |
---|
575 | vec x ( dim ); |
---|
576 | vec pom; |
---|
577 | int i; |
---|
578 | |
---|
579 | for ( i=0;i<N;i++ ) { |
---|
580 | NorRNG.sample_vector ( dim,x ); |
---|
581 | pom = R.sqrt_mult ( x ); |
---|
582 | pom +=mu; |
---|
583 | X.set_col ( i, pom ); |
---|
584 | } |
---|
585 | |
---|
586 | return X; |
---|
587 | }; |
---|
588 | |
---|
589 | template<class sq_T> |
---|
590 | double enorm<sq_T>::eval ( const vec &val ) const { |
---|
591 | double pdfl,e; |
---|
592 | pdfl = evalpdflog ( val ); |
---|
593 | e = exp ( pdfl ); |
---|
594 | return e; |
---|
595 | }; |
---|
596 | |
---|
597 | template<class sq_T> |
---|
598 | double enorm<sq_T>::evalpdflog_nn ( const vec &val ) const { |
---|
599 | // 1.83787706640935 = log(2pi) |
---|
600 | double tmp=-0.5* ( R.invqform ( mu-val ) );// - lognc(); |
---|
601 | return tmp; |
---|
602 | }; |
---|
603 | |
---|
604 | template<class sq_T> |
---|
605 | inline double enorm<sq_T>::lognc () const { |
---|
606 | // 1.83787706640935 = log(2pi) |
---|
607 | double tmp=0.5* ( R.cols() * 1.83787706640935 +R.logdet() ); |
---|
608 | return tmp; |
---|
609 | }; |
---|
610 | |
---|
611 | template<class sq_T> |
---|
612 | mlnorm<sq_T>::mlnorm ( const RV &rv0, const RV &rvc0 ) :mEF ( rv0,rvc0 ),epdf ( rv0 ),A ( rv0.count(),rv0.count() ),_mu ( epdf._mu() ) { |
---|
613 | ep =&epdf; |
---|
614 | } |
---|
615 | |
---|
616 | template<class sq_T> |
---|
617 | void mlnorm<sq_T>::set_parameters ( const mat &A0, const vec &mu0, const sq_T &R0 ) { |
---|
618 | epdf.set_parameters ( zeros ( rv.count() ),R0 ); |
---|
619 | A = A0; |
---|
620 | mu_const = mu0; |
---|
621 | } |
---|
622 | |
---|
623 | // template<class sq_T> |
---|
624 | // vec mlnorm<sq_T>::samplecond (const vec &cond, double &lik ) { |
---|
625 | // this->condition ( cond ); |
---|
626 | // vec smp = epdf.sample(); |
---|
627 | // lik = epdf.eval ( smp ); |
---|
628 | // return smp; |
---|
629 | // } |
---|
630 | |
---|
631 | // template<class sq_T> |
---|
632 | // mat mlnorm<sq_T>::samplecond (const vec &cond, vec &lik, int n ) { |
---|
633 | // int i; |
---|
634 | // int dim = rv.count(); |
---|
635 | // mat Smp ( dim,n ); |
---|
636 | // vec smp ( dim ); |
---|
637 | // this->condition ( cond ); |
---|
638 | // |
---|
639 | // for ( i=0; i<n; i++ ) { |
---|
640 | // smp = epdf.sample(); |
---|
641 | // lik ( i ) = epdf.eval ( smp ); |
---|
642 | // Smp.set_col ( i ,smp ); |
---|
643 | // } |
---|
644 | // |
---|
645 | // return Smp; |
---|
646 | // } |
---|
647 | |
---|
648 | template<class sq_T> |
---|
649 | void mlnorm<sq_T>::condition ( const vec &cond ) { |
---|
650 | _mu = A*cond + mu_const; |
---|
651 | //R is already assigned; |
---|
652 | } |
---|
653 | |
---|
654 | template<class sq_T> |
---|
655 | epdf* enorm<sq_T>::marginal ( const RV &rvn ) const { |
---|
656 | ivec irvn = rvn.dataind ( rv ); |
---|
657 | |
---|
658 | sq_T Rn ( R,irvn ); |
---|
659 | enorm<sq_T>* tmp = new enorm<sq_T> ( rvn ); |
---|
660 | tmp->set_parameters ( mu ( irvn ), Rn ); |
---|
661 | return tmp; |
---|
662 | } |
---|
663 | |
---|
664 | template<class sq_T> |
---|
665 | mpdf* enorm<sq_T>::condition ( const RV &rvn ) const { |
---|
666 | |
---|
667 | RV rvc = rv.subt ( rvn ); |
---|
668 | it_assert_debug ( ( rvc.count() +rvn.count() ==rv.count() ),"wrong rvn" ); |
---|
669 | //Permutation vector of the new R |
---|
670 | ivec irvn = rvn.dataind ( rv ); |
---|
671 | ivec irvc = rvc.dataind ( rv ); |
---|
672 | ivec perm=concat ( irvn , irvc ); |
---|
673 | sq_T Rn ( R,perm ); |
---|
674 | |
---|
675 | //fixme - could this be done in general for all sq_T? |
---|
676 | mat S=Rn.to_mat(); |
---|
677 | //fixme |
---|
678 | int n=rvn.count()-1; |
---|
679 | int end=R.rows()-1; |
---|
680 | mat S11 = S.get ( 0,n, 0, n ); |
---|
681 | mat S12 = S.get ( 0, n , rvn.count(), end ); |
---|
682 | mat S22 = S.get ( rvn.count(), end, rvn.count(), end ); |
---|
683 | |
---|
684 | vec mu1 = mu ( irvn ); |
---|
685 | vec mu2 = mu ( irvc ); |
---|
686 | mat A=S12*inv ( S22 ); |
---|
687 | sq_T R_n ( S11 - A *S12.T() ); |
---|
688 | |
---|
689 | mlnorm<sq_T>* tmp=new mlnorm<sq_T> ( rvn,rvc ); |
---|
690 | |
---|
691 | tmp->set_parameters ( A,mu1-A*mu2,R_n ); |
---|
692 | return tmp; |
---|
693 | } |
---|
694 | |
---|
695 | /////////// |
---|
696 | |
---|
697 | template<class sq_T> |
---|
698 | std::ostream &operator<< ( std::ostream &os, mlnorm<sq_T> &ml ) { |
---|
699 | os << "A:"<< ml.A<<endl; |
---|
700 | os << "mu:"<< ml.mu_const<<endl; |
---|
701 | os << "R:" << ml.epdf._R().to_mat() <<endl; |
---|
702 | return os; |
---|
703 | }; |
---|
704 | |
---|
705 | #endif //EF_H |
---|