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 {return evalpdflog_nn ( val )-lognc();} |
---|
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 | |
---|
98 | template<class sq_T> |
---|
99 | class mlnorm; |
---|
100 | |
---|
101 | /*! |
---|
102 | * \brief Gaussian density with positive definite (decomposed) covariance matrix. |
---|
103 | |
---|
104 | * More?... |
---|
105 | */ |
---|
106 | template<class sq_T> |
---|
107 | class enorm : public eEF { |
---|
108 | protected: |
---|
109 | //! mean value |
---|
110 | vec mu; |
---|
111 | //! Covariance matrix in decomposed form |
---|
112 | sq_T R; |
---|
113 | //! dimension (redundant from rv.count() for easier coding ) |
---|
114 | int dim; |
---|
115 | public: |
---|
116 | //!Default constructor |
---|
117 | enorm ( const RV &rv ); |
---|
118 | //! Set mean value \c mu and covariance \c R |
---|
119 | void set_parameters ( const vec &mu,const sq_T &R ); |
---|
120 | ////! tupdate in exponential form (not really handy) |
---|
121 | //void tupdate ( double phi, mat &vbar, double nubar ); |
---|
122 | //! dupdate in exponential form (not really handy) |
---|
123 | void dupdate ( mat &v,double nu=1.0 ); |
---|
124 | |
---|
125 | vec sample() const; |
---|
126 | //! TODO is it used? |
---|
127 | mat sample ( int N ) const; |
---|
128 | double eval ( const vec &val ) const ; |
---|
129 | double evalpdflog_nn ( const vec &val ) const; |
---|
130 | double lognc () const; |
---|
131 | vec mean() const {return mu;} |
---|
132 | mlnorm<sq_T>* condition ( const RV &rvn ); |
---|
133 | enorm<sq_T>* marginal ( const RV &rv ); |
---|
134 | //Access methods |
---|
135 | //! returns a pointer to the internal mean value. Use with Care! |
---|
136 | vec& _mu() {return mu;} |
---|
137 | |
---|
138 | //! access function |
---|
139 | void set_mu ( const vec mu0 ) { mu=mu0;} |
---|
140 | |
---|
141 | //! returns pointers to the internal variance and its inverse. Use with Care! |
---|
142 | sq_T& _R() {return R;} |
---|
143 | |
---|
144 | //! access method |
---|
145 | // mat getR () {return R.to_mat();} |
---|
146 | }; |
---|
147 | |
---|
148 | /*! |
---|
149 | * \brief Gauss-inverse-Wishart density stored in LD form |
---|
150 | |
---|
151 | * For \f$p\f$-variate densities, given rv.count() should be \f$p\times\f$ V.rows(). |
---|
152 | * |
---|
153 | */ |
---|
154 | class egiw : public eEF { |
---|
155 | protected: |
---|
156 | //! Extended information matrix of sufficient statistics |
---|
157 | ldmat V; |
---|
158 | //! Number of data records (degrees of freedom) of sufficient statistics |
---|
159 | double nu; |
---|
160 | //! Dimension of the output |
---|
161 | int xdim; |
---|
162 | //! Dimension of the regressor |
---|
163 | int nPsi; |
---|
164 | public: |
---|
165 | //!Default constructor, assuming |
---|
166 | egiw ( RV rv, mat V0, double nu0 ) : eEF ( rv ), V ( V0 ), nu ( nu0 ) { |
---|
167 | xdim = rv.count() /V.rows(); |
---|
168 | it_assert_debug ( rv.count() ==xdim*V.rows(),"Incompatible V0." ); |
---|
169 | nPsi = V.rows()-xdim; |
---|
170 | } |
---|
171 | //!Full constructor for V in ldmat form |
---|
172 | egiw ( RV rv, ldmat V0, double nu0 ) : eEF ( rv ), V ( V0 ), nu ( nu0 ) { |
---|
173 | xdim = rv.count() /V.rows(); |
---|
174 | it_assert_debug ( rv.count() ==xdim*V.rows(),"Incompatible V0." ); |
---|
175 | nPsi = V.rows()-xdim; |
---|
176 | } |
---|
177 | |
---|
178 | vec sample() const; |
---|
179 | vec mean() const; |
---|
180 | void mean_mat ( mat &M, mat&R ) const; |
---|
181 | //! In this instance, val= [theta, r]. For multivariate instances, it is stored columnwise val = [theta_1 theta_2 ... r_1 r_2 ] |
---|
182 | double evalpdflog_nn ( const vec &val ) const; |
---|
183 | double lognc () const; |
---|
184 | |
---|
185 | //Access |
---|
186 | //! returns a pointer to the internal statistics. Use with Care! |
---|
187 | ldmat& _V() {return V;} |
---|
188 | //! returns a pointer to the internal statistics. Use with Care! |
---|
189 | double& _nu() {return nu;} |
---|
190 | void pow ( double p ); |
---|
191 | }; |
---|
192 | |
---|
193 | /*! \brief Dirichlet posterior density |
---|
194 | |
---|
195 | Continuous Dirichlet density of \f$n\f$-dimensional variable \f$x\f$ |
---|
196 | \f[ |
---|
197 | f(x|\beta) = \frac{\Gamma[\gamma]}{\prod_{i=1}^{n}\Gamma(\beta_i)} \prod_{i=1}^{n}x_i^{\beta_i-1} |
---|
198 | \f] |
---|
199 | where \f$\gamma=\sum_i \beta_i\f$. |
---|
200 | */ |
---|
201 | class eDirich: public eEF { |
---|
202 | protected: |
---|
203 | //!sufficient statistics |
---|
204 | vec beta; |
---|
205 | public: |
---|
206 | //!Default constructor |
---|
207 | eDirich ( const RV &rv, const vec &beta0 ) : eEF ( rv ),beta ( beta0 ) {it_assert_debug ( rv.count() ==beta.length(),"Incompatible statistics" ); }; |
---|
208 | //! Copy constructor |
---|
209 | eDirich ( const eDirich &D0 ) : eEF ( D0.rv ),beta ( D0.beta ) {}; |
---|
210 | vec sample() const {it_error ( "Not implemented" );return vec_1 ( 0.0 );}; |
---|
211 | vec mean() const {return beta/sum ( beta );}; |
---|
212 | //! In this instance, val is ... |
---|
213 | double evalpdflog_nn ( const vec &val ) const {return ( beta-1 ) *log ( val );}; |
---|
214 | double lognc () const { |
---|
215 | double gam=sum ( beta ); |
---|
216 | double lgb=0.0; |
---|
217 | for ( int i=0;i<beta.length();i++ ) {lgb+=lgamma ( beta ( i ) );} |
---|
218 | return lgb-lgamma ( gam ); |
---|
219 | }; |
---|
220 | //!access function |
---|
221 | vec& _beta() {return beta;} |
---|
222 | //!Set internal parameters |
---|
223 | void set_parameters ( const vec &beta0 ) { |
---|
224 | if ( beta0.length() !=beta.length() ) { |
---|
225 | it_assert_debug ( rv.length() ==1,"Undefined" ); |
---|
226 | rv.set_size ( 0,beta0.length() ); |
---|
227 | } |
---|
228 | beta= beta0; |
---|
229 | } |
---|
230 | }; |
---|
231 | |
---|
232 | //! Estimator for Multinomial density |
---|
233 | class multiBM : public BMEF { |
---|
234 | protected: |
---|
235 | //! Conjugate prior and posterior |
---|
236 | eDirich est; |
---|
237 | vec β |
---|
238 | public: |
---|
239 | //!Default constructor |
---|
240 | multiBM ( const RV &rv, const vec beta0 ) : BMEF ( rv ),est ( rv,beta0 ),beta ( est._beta() ) {last_lognc=est.lognc();} |
---|
241 | //!Copy constructor |
---|
242 | multiBM ( const multiBM &B ) : BMEF ( B ),est ( rv,B.beta ),beta ( est._beta() ) {} |
---|
243 | |
---|
244 | void set_statistics ( const BM* mB0 ) {const multiBM* mB=dynamic_cast<const multiBM*> ( mB0 ); beta=mB->beta;} |
---|
245 | void bayes ( const vec &dt ) { |
---|
246 | if ( frg<1.0 ) {beta*=frg;last_lognc=est.lognc();} |
---|
247 | beta+=dt; |
---|
248 | if ( evalll ) {ll=est.lognc()-last_lognc;} |
---|
249 | } |
---|
250 | double logpred ( const vec &dt ) const { |
---|
251 | eDirich pred ( est ); |
---|
252 | vec &beta = pred._beta(); |
---|
253 | |
---|
254 | double lll; |
---|
255 | if ( frg<1.0 ) |
---|
256 | {beta*=frg;lll=pred.lognc();} |
---|
257 | else |
---|
258 | if ( evalll ) {lll=last_lognc;} |
---|
259 | else{lll=pred.lognc();} |
---|
260 | |
---|
261 | beta+=dt; |
---|
262 | return pred.lognc()-lll; |
---|
263 | } |
---|
264 | void flatten ( const BMEF* B ) { |
---|
265 | const eDirich* E=dynamic_cast<const eDirich*> ( B ); |
---|
266 | // sum(beta) should be equal to sum(B.beta) |
---|
267 | const vec &Eb=const_cast<eDirich*> ( E )->_beta(); |
---|
268 | est.pow ( sum ( beta ) /sum ( Eb ) ); |
---|
269 | if ( evalll ) {last_lognc=est.lognc();} |
---|
270 | } |
---|
271 | const epdf& _epdf() const {return est;}; |
---|
272 | void set_parameters ( const vec &beta0 ) { |
---|
273 | est.set_parameters ( beta0 ); |
---|
274 | rv = est._rv(); |
---|
275 | if ( evalll ) {last_lognc=est.lognc();} |
---|
276 | } |
---|
277 | }; |
---|
278 | |
---|
279 | /*! |
---|
280 | \brief Gamma posterior density |
---|
281 | |
---|
282 | Multivariate Gamma density as product of independent univariate densities. |
---|
283 | \f[ |
---|
284 | f(x|\alpha,\beta) = \prod f(x_i|\alpha_i,\beta_i) |
---|
285 | \f] |
---|
286 | */ |
---|
287 | |
---|
288 | class egamma : public eEF { |
---|
289 | protected: |
---|
290 | //! Vector \f$\alpha\f$ |
---|
291 | vec alpha; |
---|
292 | //! Vector \f$\beta\f$ |
---|
293 | vec beta; |
---|
294 | public : |
---|
295 | //! Default constructor |
---|
296 | egamma ( const RV &rv ) :eEF ( rv ) {}; |
---|
297 | //! Sets parameters |
---|
298 | void set_parameters ( const vec &a, const vec &b ) {alpha=a,beta=b;}; |
---|
299 | vec sample() const; |
---|
300 | //! TODO: is it used anywhere? |
---|
301 | // mat sample ( int N ) const; |
---|
302 | double evalpdflog ( const vec &val ) const; |
---|
303 | double lognc () const; |
---|
304 | //! Returns poiter to alpha and beta. Potentially dengerous: use with care! |
---|
305 | void _param ( vec* &a, vec* &b ) {a=αb=β}; |
---|
306 | vec mean() const {vec pom ( alpha ); pom/=beta; return pom;} |
---|
307 | }; |
---|
308 | /* |
---|
309 | //! Weighted mixture of epdfs with external owned components. |
---|
310 | class emix : public epdf { |
---|
311 | protected: |
---|
312 | int n; |
---|
313 | vec &w; |
---|
314 | Array<epdf*> Coms; |
---|
315 | public: |
---|
316 | //! Default constructor |
---|
317 | emix ( const RV &rv, vec &w0): epdf(rv), n(w0.length()), w(w0), Coms(n) {}; |
---|
318 | void set_parameters( int &i, double wi, epdf* ep){w(i)=wi;Coms(i)=ep;} |
---|
319 | vec mean(){vec pom; for(int i=0;i<n;i++){pom+=Coms(i)->mean()*w(i);} return pom;}; |
---|
320 | vec sample() {it_error ( "Not implemented" );return 0;} |
---|
321 | }; |
---|
322 | */ |
---|
323 | |
---|
324 | //! Uniform distributed density on a rectangular support |
---|
325 | |
---|
326 | class euni: public epdf { |
---|
327 | protected: |
---|
328 | //! lower bound on support |
---|
329 | vec low; |
---|
330 | //! upper bound on support |
---|
331 | vec high; |
---|
332 | //! internal |
---|
333 | vec distance; |
---|
334 | //! normalizing coefficients |
---|
335 | double nk; |
---|
336 | //! cache of log( \c nk ) |
---|
337 | double lnk; |
---|
338 | public: |
---|
339 | //! Defualt constructor |
---|
340 | euni ( const RV rv ) :epdf ( rv ) {} |
---|
341 | double eval ( const vec &val ) const {return nk;} |
---|
342 | double evalpdflog ( const vec &val ) const {return lnk;} |
---|
343 | vec sample() const { |
---|
344 | vec smp ( rv.count() ); |
---|
345 | #pragma omp critical |
---|
346 | UniRNG.sample_vector ( rv.count(),smp ); |
---|
347 | return low+elem_mult ( distance,smp ); |
---|
348 | } |
---|
349 | //! set values of \c low and \c high |
---|
350 | void set_parameters ( const vec &low0, const vec &high0 ) { |
---|
351 | distance = high0-low0; |
---|
352 | it_assert_debug ( min ( distance ) >0.0,"bad support" ); |
---|
353 | low = low0; |
---|
354 | high = high0; |
---|
355 | nk = prod ( 1.0/distance ); |
---|
356 | lnk = log ( nk ); |
---|
357 | } |
---|
358 | vec mean() const {vec pom=high; pom-=low; pom/=2.0; return pom;} |
---|
359 | }; |
---|
360 | |
---|
361 | |
---|
362 | /*! |
---|
363 | \brief Normal distributed linear function with linear function of mean value; |
---|
364 | |
---|
365 | Mean value \f$mu=A*rvc+mu_0\f$. |
---|
366 | */ |
---|
367 | template<class sq_T> |
---|
368 | class mlnorm : public mEF { |
---|
369 | //! Internal epdf that arise by conditioning on \c rvc |
---|
370 | enorm<sq_T> epdf; |
---|
371 | mat A; |
---|
372 | vec mu_const; |
---|
373 | vec& _mu; //cached epdf.mu; |
---|
374 | public: |
---|
375 | //! Constructor |
---|
376 | mlnorm (const RV &rv, const RV &rvc ); |
---|
377 | //! Set \c A and \c R |
---|
378 | void set_parameters ( const mat &A, const vec &mu0, const sq_T &R ); |
---|
379 | //!Generate one sample of the posterior |
---|
380 | vec samplecond (const vec &cond, double &lik ); |
---|
381 | //!Generate matrix of samples of the posterior |
---|
382 | mat samplecond (const vec &cond, vec &lik, int n ); |
---|
383 | //! Set value of \c rvc . Result of this operation is stored in \c epdf use function \c _ep to access it. |
---|
384 | void condition (const vec &cond ); |
---|
385 | }; |
---|
386 | |
---|
387 | /*! |
---|
388 | \brief Gamma random walk |
---|
389 | |
---|
390 | Mean value, \f$\mu\f$, of this density is given by \c rvc . |
---|
391 | Standard deviation of the random walk is proportional to one \f$k\f$-th the mean. |
---|
392 | This is achieved by setting \f$\alpha=k\f$ and \f$\beta=k/\mu\f$. |
---|
393 | |
---|
394 | The standard deviation of the walk is then: \f$\mu/\sqrt(k)\f$. |
---|
395 | */ |
---|
396 | class mgamma : public mEF { |
---|
397 | protected: |
---|
398 | //! Internal epdf that arise by conditioning on \c rvc |
---|
399 | egamma epdf; |
---|
400 | //! Constant \f$k\f$ |
---|
401 | double k; |
---|
402 | //! cache of epdf.beta |
---|
403 | vec* _beta; |
---|
404 | |
---|
405 | public: |
---|
406 | //! Constructor |
---|
407 | mgamma ( const RV &rv,const RV &rvc ); |
---|
408 | //! Set value of \c k |
---|
409 | void set_parameters ( double k ); |
---|
410 | void condition ( const vec &val ) {*_beta=k/val;}; |
---|
411 | }; |
---|
412 | |
---|
413 | /*! |
---|
414 | \brief Gamma random walk around a fixed point |
---|
415 | |
---|
416 | 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 |
---|
417 | \f[ \mu = \mu_{t-1} ^{l} p^{1-l}\f] |
---|
418 | |
---|
419 | Standard deviation of the random walk is proportional to one \f$k\f$-th the mean. |
---|
420 | This is achieved by setting \f$\alpha=k\f$ and \f$\beta=k/\mu\f$. |
---|
421 | |
---|
422 | The standard deviation of the walk is then: \f$\mu/\sqrt(k)\f$. |
---|
423 | */ |
---|
424 | class mgamma_fix : public mgamma { |
---|
425 | protected: |
---|
426 | //! parameter l |
---|
427 | double l; |
---|
428 | //! reference vector |
---|
429 | vec refl; |
---|
430 | public: |
---|
431 | //! Constructor |
---|
432 | mgamma_fix ( const RV &rv,const RV &rvc ) : mgamma ( rv,rvc ),refl ( rv.count() ) {}; |
---|
433 | //! Set value of \c k |
---|
434 | void set_parameters ( double k0 , vec ref0, double l0 ) { |
---|
435 | mgamma::set_parameters ( k0 ); |
---|
436 | refl=pow ( ref0,1.0-l0 );l=l0; |
---|
437 | }; |
---|
438 | |
---|
439 | void condition ( const vec &val ) {vec mean=elem_mult ( refl,pow ( val,l ) ); *_beta=k/mean;}; |
---|
440 | }; |
---|
441 | |
---|
442 | //! Switch between various resampling methods. |
---|
443 | enum RESAMPLING_METHOD { MULTINOMIAL = 0, STRATIFIED = 1, SYSTEMATIC = 3 }; |
---|
444 | /*! |
---|
445 | \brief Weighted empirical density |
---|
446 | |
---|
447 | Used e.g. in particle filters. |
---|
448 | */ |
---|
449 | class eEmp: public epdf { |
---|
450 | protected : |
---|
451 | //! Number of particles |
---|
452 | int n; |
---|
453 | //! Sample weights \f$w\f$ |
---|
454 | vec w; |
---|
455 | //! Samples \f$x^{(i)}, i=1..n\f$ |
---|
456 | Array<vec> samples; |
---|
457 | public: |
---|
458 | //! Default constructor |
---|
459 | eEmp ( const RV &rv0 ,int n0 ) :epdf ( rv0 ),n ( n0 ),w ( n ),samples ( n ) {}; |
---|
460 | //! Set samples and weights |
---|
461 | void set_parameters ( const vec &w0, const epdf* pdf0 ); |
---|
462 | //! Set sample |
---|
463 | void set_samples ( const epdf* pdf0 ); |
---|
464 | //! Potentially dangerous, use with care. |
---|
465 | vec& _w() {return w;}; |
---|
466 | //! access function |
---|
467 | Array<vec>& _samples() {return samples;}; |
---|
468 | //! 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. |
---|
469 | ivec resample ( RESAMPLING_METHOD method = SYSTEMATIC ); |
---|
470 | //! inherited operation : NOT implemneted |
---|
471 | vec sample() const {it_error ( "Not implemented" );return 0;} |
---|
472 | //! inherited operation : NOT implemneted |
---|
473 | double evalpdflog ( const vec &val ) const {it_error ( "Not implemented" );return 0.0;} |
---|
474 | vec mean() const { |
---|
475 | vec pom=zeros ( rv.count() ); |
---|
476 | for ( int i=0;i<n;i++ ) {pom+=samples ( i ) *w ( i );} |
---|
477 | return pom; |
---|
478 | } |
---|
479 | }; |
---|
480 | |
---|
481 | |
---|
482 | //////////////////////// |
---|
483 | |
---|
484 | template<class sq_T> |
---|
485 | enorm<sq_T>::enorm ( const RV &rv ) :eEF ( rv ), mu ( rv.count() ),R ( rv.count() ),dim ( rv.count() ) {}; |
---|
486 | |
---|
487 | template<class sq_T> |
---|
488 | void enorm<sq_T>::set_parameters ( const vec &mu0, const sq_T &R0 ) { |
---|
489 | //Fixme test dimensions of mu0 and R0; |
---|
490 | mu = mu0; |
---|
491 | R = R0; |
---|
492 | }; |
---|
493 | |
---|
494 | template<class sq_T> |
---|
495 | void enorm<sq_T>::dupdate ( mat &v, double nu ) { |
---|
496 | // |
---|
497 | }; |
---|
498 | |
---|
499 | // template<class sq_T> |
---|
500 | // void enorm<sq_T>::tupdate ( double phi, mat &vbar, double nubar ) { |
---|
501 | // // |
---|
502 | // }; |
---|
503 | |
---|
504 | template<class sq_T> |
---|
505 | vec enorm<sq_T>::sample() const { |
---|
506 | vec x ( dim ); |
---|
507 | NorRNG.sample_vector ( dim,x ); |
---|
508 | vec smp = R.sqrt_mult ( x ); |
---|
509 | |
---|
510 | smp += mu; |
---|
511 | return smp; |
---|
512 | }; |
---|
513 | |
---|
514 | template<class sq_T> |
---|
515 | mat enorm<sq_T>::sample ( int N ) const { |
---|
516 | mat X ( dim,N ); |
---|
517 | vec x ( dim ); |
---|
518 | vec pom; |
---|
519 | int i; |
---|
520 | |
---|
521 | for ( i=0;i<N;i++ ) { |
---|
522 | NorRNG.sample_vector ( dim,x ); |
---|
523 | pom = R.sqrt_mult ( x ); |
---|
524 | pom +=mu; |
---|
525 | X.set_col ( i, pom ); |
---|
526 | } |
---|
527 | |
---|
528 | return X; |
---|
529 | }; |
---|
530 | |
---|
531 | template<class sq_T> |
---|
532 | double enorm<sq_T>::eval ( const vec &val ) const { |
---|
533 | double pdfl,e; |
---|
534 | pdfl = evalpdflog ( val ); |
---|
535 | e = exp ( pdfl ); |
---|
536 | return e; |
---|
537 | }; |
---|
538 | |
---|
539 | template<class sq_T> |
---|
540 | double enorm<sq_T>::evalpdflog_nn ( const vec &val ) const { |
---|
541 | // 1.83787706640935 = log(2pi) |
---|
542 | return -0.5* ( R.invqform ( mu-val ) );// - lognc(); |
---|
543 | }; |
---|
544 | |
---|
545 | template<class sq_T> |
---|
546 | inline double enorm<sq_T>::lognc () const { |
---|
547 | // 1.83787706640935 = log(2pi) |
---|
548 | return 0.5* ( R.cols() * 1.83787706640935 +R.logdet() ); |
---|
549 | }; |
---|
550 | |
---|
551 | template<class sq_T> |
---|
552 | mlnorm<sq_T>::mlnorm ( const RV &rv0, const RV &rvc0 ) :mEF ( rv0,rvc0 ),epdf ( rv0 ),A ( rv0.count(),rv0.count() ),_mu ( epdf._mu() ) { |
---|
553 | ep =&epdf; |
---|
554 | } |
---|
555 | |
---|
556 | template<class sq_T> |
---|
557 | void mlnorm<sq_T>::set_parameters ( const mat &A0, const vec &mu0, const sq_T &R0 ) { |
---|
558 | epdf.set_parameters ( zeros ( rv.count() ),R0 ); |
---|
559 | A = A0; |
---|
560 | mu_const = mu0; |
---|
561 | } |
---|
562 | |
---|
563 | template<class sq_T> |
---|
564 | vec mlnorm<sq_T>::samplecond (const vec &cond, double &lik ) { |
---|
565 | this->condition ( cond ); |
---|
566 | vec smp = epdf.sample(); |
---|
567 | lik = epdf.eval ( smp ); |
---|
568 | return smp; |
---|
569 | } |
---|
570 | |
---|
571 | template<class sq_T> |
---|
572 | mat mlnorm<sq_T>::samplecond (const vec &cond, vec &lik, int n ) { |
---|
573 | int i; |
---|
574 | int dim = rv.count(); |
---|
575 | mat Smp ( dim,n ); |
---|
576 | vec smp ( dim ); |
---|
577 | this->condition ( cond ); |
---|
578 | |
---|
579 | for ( i=0; i<n; i++ ) { |
---|
580 | smp = epdf.sample(); |
---|
581 | lik ( i ) = epdf.eval ( smp ); |
---|
582 | Smp.set_col ( i ,smp ); |
---|
583 | } |
---|
584 | |
---|
585 | return Smp; |
---|
586 | } |
---|
587 | |
---|
588 | template<class sq_T> |
---|
589 | void mlnorm<sq_T>::condition (const vec &cond ) { |
---|
590 | _mu = A*cond + mu_const; |
---|
591 | //R is already assigned; |
---|
592 | } |
---|
593 | |
---|
594 | template<class sq_T> |
---|
595 | enorm<sq_T>* enorm<sq_T>::marginal ( const RV &rvn ) { |
---|
596 | ivec irvn = rvn.dataind ( rv ); |
---|
597 | |
---|
598 | sq_T Rn ( R,irvn ); |
---|
599 | enorm<sq_T>* tmp = new enorm<sq_T>( rvn ); |
---|
600 | tmp->set_parameters ( mu ( irvn ), Rn ); |
---|
601 | return tmp; |
---|
602 | } |
---|
603 | |
---|
604 | template<class sq_T> |
---|
605 | mlnorm<sq_T>* enorm<sq_T>::condition ( const RV &rvn ) { |
---|
606 | |
---|
607 | RV rvc = rv.subt ( rvn ); |
---|
608 | it_assert_debug ( ( rvc.count() +rvn.count() ==rv.count() ),"wrong rvn" ); |
---|
609 | //Permutation vector of the new R |
---|
610 | ivec irvn = rvn.dataind ( rv ); |
---|
611 | ivec irvc = rvc.dataind ( rv ); |
---|
612 | ivec perm=concat ( irvn , irvc ); |
---|
613 | sq_T Rn ( R,perm ); |
---|
614 | |
---|
615 | //fixme - could this be done in general for all sq_T? |
---|
616 | mat S=R.to_mat(); |
---|
617 | //fixme |
---|
618 | int n=rvn.count()-1; |
---|
619 | int end=R.rows()-1; |
---|
620 | mat S11 = S.get ( 0,n, 0, n ); |
---|
621 | mat S12 = S.get ( rvn.count(), end, 0, n ); |
---|
622 | mat S22 = S.get ( rvn.count(), end, rvn.count(), end ); |
---|
623 | |
---|
624 | vec mu1 = mu ( irvn ); |
---|
625 | vec mu2 = mu ( irvc ); |
---|
626 | mat A=S12*inv ( S22 ); |
---|
627 | sq_T R_n ( S11 - A *S12.T() ); |
---|
628 | |
---|
629 | mlnorm<sq_T>* tmp=new mlnorm<sq_T> ( rvn,rvc ); |
---|
630 | |
---|
631 | tmp->set_parameters ( A,mu1-A*mu2,R_n ); |
---|
632 | return tmp; |
---|
633 | } |
---|
634 | |
---|
635 | /////////// |
---|
636 | |
---|
637 | |
---|
638 | #endif //EF_H |
---|