1 | #include <math.h> |
---|
2 | |
---|
3 | #include <itpp/base/bessel.h> |
---|
4 | #include "exp_family.h" |
---|
5 | |
---|
6 | namespace bdm { |
---|
7 | |
---|
8 | Uniform_RNG UniRNG; |
---|
9 | Normal_RNG NorRNG; |
---|
10 | Gamma_RNG GamRNG; |
---|
11 | |
---|
12 | using std::cout; |
---|
13 | |
---|
14 | /////////// |
---|
15 | |
---|
16 | void BMEF::bayes ( const vec &yt, const vec &cond ) { |
---|
17 | this->bayes_weighted ( yt, cond, 1.0 ); |
---|
18 | }; |
---|
19 | |
---|
20 | void egiw::set_parameters ( int dimx0, ldmat V0, double nu0 ) { |
---|
21 | dimx = dimx0; |
---|
22 | nPsi = V0.rows()-dimx; |
---|
23 | V = V0; |
---|
24 | if ( nu0 < 0 ) { |
---|
25 | nu = 0.1 + nPsi + 2 * dimx + 2; // +2 assures finite expected value of R |
---|
26 | // terms before that are sufficient for finite normalization |
---|
27 | } else { |
---|
28 | nu = nu0; |
---|
29 | } |
---|
30 | } |
---|
31 | |
---|
32 | vec egiw::sample() const { |
---|
33 | mat M; |
---|
34 | chmat R; |
---|
35 | sample_mat ( M, R ); |
---|
36 | |
---|
37 | return concat ( cvectorize ( M ), cvectorize ( R.to_mat() ) ); |
---|
38 | } |
---|
39 | |
---|
40 | mat egiw::sample_mat ( int n ) const { |
---|
41 | // TODO - correct approach - convert to product of norm * Wishart |
---|
42 | mat M; |
---|
43 | ldmat Vz; |
---|
44 | ldmat Lam; |
---|
45 | factorize ( M, Vz, Lam ); |
---|
46 | |
---|
47 | chmat ChLam ( Lam.to_mat() ); |
---|
48 | chmat iChLam; |
---|
49 | ChLam.inv ( iChLam ); |
---|
50 | |
---|
51 | eWishartCh Omega; //inverse Wishart, result is R, |
---|
52 | Omega.set_parameters ( iChLam, nu - 2*nPsi - dimx ); // 2*nPsi is there to match numercial simulations - check if analytically correct |
---|
53 | Omega.validate(); |
---|
54 | |
---|
55 | mat OmChi; |
---|
56 | mat Z ( M.rows(), M.cols() ); |
---|
57 | |
---|
58 | mat Mi; |
---|
59 | mat RChiT; |
---|
60 | mat tmp ( dimension(), n ); |
---|
61 | M=M.T();// ugly hack == decide what to do with M. |
---|
62 | for ( int i = 0; i < n; i++ ) { |
---|
63 | OmChi = Omega.sample_mat(); |
---|
64 | RChiT = inv ( OmChi ); |
---|
65 | Z = randn ( M.rows(), M.cols() ); |
---|
66 | Mi = M + RChiT * Z * inv ( Vz._L().T() * diag ( sqrt ( Vz._D() ) ) ); |
---|
67 | |
---|
68 | tmp.set_col ( i, concat ( cvectorize ( Mi ), cvectorize ( RChiT*RChiT.T() ) ) ); |
---|
69 | } |
---|
70 | return tmp; |
---|
71 | } |
---|
72 | |
---|
73 | void egiw::sample_mat ( mat &Mi, chmat &Ri ) const { |
---|
74 | |
---|
75 | // TODO - correct approach - convert to product of norm * Wishart |
---|
76 | mat M; |
---|
77 | ldmat Vz; |
---|
78 | ldmat Lam; |
---|
79 | factorize ( M, Vz, Lam ); |
---|
80 | |
---|
81 | chmat Ch; |
---|
82 | Ch.setCh ( Lam._L() *diag ( sqrt ( Lam._D() ) ) ); |
---|
83 | chmat iCh; |
---|
84 | Ch.inv ( iCh ); |
---|
85 | |
---|
86 | eWishartCh Omega; //inverse Wishart, result is R, |
---|
87 | Omega.set_parameters ( iCh, nu - 2*nPsi - dimx ); // 2*nPsi is there to match numercial simulations - check if analytically correct |
---|
88 | Omega.validate(); |
---|
89 | |
---|
90 | chmat Omi; |
---|
91 | Omi.setCh ( Omega.sample_mat() ); |
---|
92 | |
---|
93 | if (M._datasize()>0) { |
---|
94 | mat Z = randn ( M.rows(), M.cols() ); |
---|
95 | Mi = M + Omi._Ch() * Z * inv ( Vz._L() * diag ( sqrt ( Vz._D() ) ) ); |
---|
96 | } |
---|
97 | Omi.inv ( Ri ); |
---|
98 | } |
---|
99 | |
---|
100 | double egiw::evallog_nn ( const vec &val ) const { |
---|
101 | bdm_assert_debug(val.length()==dimx*(nPsi+dimx),"Incorrect cond in egiw::evallog_nn" ); |
---|
102 | |
---|
103 | int vend = val.length() - 1; |
---|
104 | |
---|
105 | if ( dimx == 1 ) { //same as the following, just quicker. |
---|
106 | double r = val ( vend ); //last entry! |
---|
107 | if ( r < 0 ) return -inf; |
---|
108 | vec Psi ( nPsi + dimx ); |
---|
109 | Psi ( 0 ) = -1.0; |
---|
110 | Psi.set_subvector ( 1, val ( 0, vend - 1 ) ); // fill the rest |
---|
111 | |
---|
112 | double Vq = V.qform ( Psi ); |
---|
113 | return -0.5* ( nu*log ( r ) + Vq / r ); |
---|
114 | } else { |
---|
115 | mat Tmp; |
---|
116 | if (nPsi>0) { |
---|
117 | mat Th = reshape ( val ( 0, nPsi * dimx - 1 ), nPsi, dimx ); |
---|
118 | Tmp = concat_vertical ( -eye ( dimx ), Th ); |
---|
119 | } else { |
---|
120 | Tmp = -eye(dimx); |
---|
121 | } |
---|
122 | fsqmat R ( reshape ( val ( nPsi*dimx, vend ), dimx, dimx ) ); |
---|
123 | double ldetR = R.logdet(); |
---|
124 | if ( !std::isfinite(ldetR) ) return -inf; |
---|
125 | fsqmat iR ( dimx ); |
---|
126 | R.inv ( iR ); |
---|
127 | |
---|
128 | return -0.5* ( nu*ldetR + trace ( iR.to_mat() *Tmp.T() *V.to_mat() *Tmp ) ); |
---|
129 | } |
---|
130 | } |
---|
131 | |
---|
132 | double egiw::lognc() const { |
---|
133 | const vec& D = V._D(); |
---|
134 | |
---|
135 | double m = nu - nPsi - dimx - 1; |
---|
136 | #define log2 0.693147180559945286226763983 |
---|
137 | #define logpi 1.144729885849400163877476189 |
---|
138 | #define log2pi 1.83787706640935 |
---|
139 | #define Inf std::numeric_limits<double>::infinity() |
---|
140 | |
---|
141 | double nkG = 0.5 * dimx * ( -nPsi * log2pi + sum ( log ( D.mid ( dimx, nPsi ) ) ) ); |
---|
142 | // temporary for lgamma in Wishart |
---|
143 | double lg = 0; |
---|
144 | for ( int i = 0; i < dimx; i++ ) { |
---|
145 | lg += lgamma ( 0.5 * ( m - i ) ); |
---|
146 | } |
---|
147 | |
---|
148 | double nkW = 0.5 * ( m * sum ( log ( D ( 0, dimx - 1 ) ) ) ) \ |
---|
149 | - 0.5 * dimx * ( m * log2 + 0.5 * ( dimx - 1 ) * log2pi ) - lg; |
---|
150 | |
---|
151 | // bdm_assert_debug ( ( ( -nkG - nkW ) > -Inf ) && ( ( -nkG - nkW ) < Inf ), "ARX improper" ); |
---|
152 | if ( -nkG - nkW == Inf ) { |
---|
153 | cout << "??" << endl; |
---|
154 | } |
---|
155 | return -nkG - nkW; |
---|
156 | } |
---|
157 | |
---|
158 | vec egiw::est_theta() const { |
---|
159 | if ( dimx == 1 ) { |
---|
160 | const mat &L = V._L(); |
---|
161 | int end = L.rows() - 1; |
---|
162 | |
---|
163 | mat iLsub = ltuinv ( L ( dimx, end, dimx, end ) ); |
---|
164 | |
---|
165 | vec L0 = L.get_col ( 0 ); |
---|
166 | |
---|
167 | return iLsub * L0 ( 1, end ); |
---|
168 | } else { |
---|
169 | bdm_error ( "ERROR: est_theta() not implemented for dimx>1" ); |
---|
170 | return vec(); |
---|
171 | } |
---|
172 | } |
---|
173 | |
---|
174 | void egiw::factorize ( mat &M, ldmat &Vz, ldmat &Lam ) const { |
---|
175 | const mat &L = V._L(); |
---|
176 | const vec &D = V._D(); |
---|
177 | int end = L.rows() - 1; |
---|
178 | |
---|
179 | Lam = ldmat ( L ( 0, dimx - 1, 0, dimx - 1 ), D ( 0, dimx - 1 ) ); //exp val of R |
---|
180 | |
---|
181 | if (dimx<=end) { |
---|
182 | Vz = ldmat ( L ( dimx, end, dimx, end ), D ( dimx, end ) ); |
---|
183 | mat iLsub = ltuinv ( Vz._L() ); |
---|
184 | // set mean value |
---|
185 | mat Lpsi = L ( dimx, end, 0, dimx - 1 ); |
---|
186 | M = iLsub * Lpsi; |
---|
187 | } |
---|
188 | /* if ( 0 ) { // test with Peterka |
---|
189 | mat VF = V.to_mat(); |
---|
190 | mat Vf = VF ( 0, dimx - 1, 0, dimx - 1 ); |
---|
191 | mat Vzf = VF ( dimx, end, 0, dimx - 1 ); |
---|
192 | mat VZ = VF ( dimx, end, dimx, end ); |
---|
193 | |
---|
194 | mat Lam2 = Vf - Vzf.T() * inv ( VZ ) * Vzf; |
---|
195 | }*/ |
---|
196 | } |
---|
197 | |
---|
198 | ldmat egiw::est_theta_cov() const { |
---|
199 | if ( dimx == 1 ) { |
---|
200 | const mat &L = V._L(); |
---|
201 | const vec &D = V._D(); |
---|
202 | int end = D.length() - 1; |
---|
203 | |
---|
204 | mat Lsub = L ( 1, end, 1, end ); |
---|
205 | // mat Dsub = diag ( D ( 1, end ) ); |
---|
206 | |
---|
207 | //ldmat LD ( inv ( Lsub ).T(), 1.0 / D ( 1, end ) ); |
---|
208 | ldmat LD; LD.ldform( inv ( Lsub ).T(), 1.0 / D ( 1, end ) ); |
---|
209 | return LD; |
---|
210 | |
---|
211 | } else { |
---|
212 | bdm_error ( "ERROR: est_theta_cov() not implemented for dimx>1" ); |
---|
213 | return ldmat(); |
---|
214 | } |
---|
215 | |
---|
216 | } |
---|
217 | |
---|
218 | vec egiw::mean() const { |
---|
219 | |
---|
220 | if ( dimx == 1 ) { |
---|
221 | const vec &D = V._D(); |
---|
222 | int end = D.length() - 1; |
---|
223 | |
---|
224 | vec m ( dim ); |
---|
225 | m.set_subvector ( 0, est_theta() ); |
---|
226 | m ( end ) = D ( 0 ) / ( nu - nPsi - 2 * dimx - 2 ); |
---|
227 | return m; |
---|
228 | } else { |
---|
229 | mat M; |
---|
230 | mat R; |
---|
231 | mean_mat ( M, R ); |
---|
232 | return concat ( cvectorize ( M ), cvectorize ( R ) ); |
---|
233 | } |
---|
234 | |
---|
235 | } |
---|
236 | |
---|
237 | vec egiw::variance() const { |
---|
238 | int l = V.rows(); |
---|
239 | // cut out rest of lower-right part of V |
---|
240 | // invert it |
---|
241 | ldmat itmp; |
---|
242 | if (dimx<l) { |
---|
243 | const ldmat tmp ( V, linspace ( dimx, l - 1 ) ); |
---|
244 | tmp.inv ( itmp ); |
---|
245 | } |
---|
246 | // following Wikipedia notation |
---|
247 | // m=nu-nPsi-dimx-1, p=dimx |
---|
248 | double mp1p = nu - nPsi - 2 * dimx; // m-p+1 |
---|
249 | double mp1m = mp1p - 2; // m-p-1 |
---|
250 | |
---|
251 | if ( dimx == 1 ) { |
---|
252 | double cove = V._D() ( 0 ) / mp1m ; |
---|
253 | |
---|
254 | vec var ( l ); |
---|
255 | var.set_subvector ( 0, diag ( itmp.to_mat() ) *cove ); |
---|
256 | var ( l - 1 ) = cove * cove / ( mp1m - 2 ); |
---|
257 | return var; |
---|
258 | } else { |
---|
259 | ldmat Vll ( V, linspace ( 0, dimx - 1 ) ); // top-left part of V |
---|
260 | mat Y = Vll.to_mat(); |
---|
261 | mat varY ( Y.rows(), Y.cols() ); |
---|
262 | |
---|
263 | double denom = ( mp1p - 1 ) * mp1m * mp1m * ( mp1m - 2 ); // (m-p)(m-p-1)^2(m-p-3) |
---|
264 | |
---|
265 | int i, j; |
---|
266 | for ( i = 0; i < Y.rows(); i++ ) { |
---|
267 | for ( j = 0; j < Y.cols(); j++ ) { |
---|
268 | varY ( i, j ) = ( mp1p * Y ( i, j ) * Y ( i, j ) + mp1m * Y ( i, i ) * Y ( j, j ) ) / denom; |
---|
269 | } |
---|
270 | } |
---|
271 | vec mean_dR = diag ( Y ) / mp1m; // corresponds to cove |
---|
272 | vec var_th = diag ( itmp.to_mat() ); |
---|
273 | vec var_Th ( mean_dR.length() *var_th.length() ); |
---|
274 | // diagonal of diag(mean_dR) \kron diag(var_th) |
---|
275 | for ( int i = 0; i < mean_dR.length(); i++ ) { |
---|
276 | var_Th.set_subvector ( i*var_th.length(), var_th*mean_dR ( i ) ); |
---|
277 | } |
---|
278 | |
---|
279 | return concat ( var_Th, cvectorize ( varY ) ); |
---|
280 | } |
---|
281 | } |
---|
282 | |
---|
283 | void egiw::mean_mat ( mat &M, mat&R ) const { |
---|
284 | const mat &L = V._L(); |
---|
285 | const vec &D = V._D(); |
---|
286 | int end = L.rows() - 1; |
---|
287 | |
---|
288 | ldmat ldR ( L ( 0, dimx - 1, 0, dimx - 1 ), D ( 0, dimx - 1 ) / ( nu - nPsi - 2*dimx - 2 ) ); //exp val of R |
---|
289 | |
---|
290 | // set mean value |
---|
291 | if (dimx<=end) { |
---|
292 | mat iLsub = ltuinv ( L ( dimx, end, dimx, end ) ); |
---|
293 | mat Lpsi = L ( dimx, end, 0, dimx - 1 ); |
---|
294 | M = iLsub * Lpsi; |
---|
295 | } |
---|
296 | R = ldR.to_mat() ; |
---|
297 | } |
---|
298 | |
---|
299 | void egiw::log_register ( bdm::logger& L, const string& prefix ) { |
---|
300 | epdf::log_register ( L, prefix ); |
---|
301 | if ( log_level[logvartheta] ) { |
---|
302 | int th_dim = dim - dimx*dimx; // dimension - dimension of cov |
---|
303 | L.add_vector( log_level, logvartheta, RV ( th_dim ), prefix ); |
---|
304 | } |
---|
305 | } |
---|
306 | |
---|
307 | void egiw::log_write() const { |
---|
308 | epdf::log_write(); |
---|
309 | if ( log_level[logvartheta] ) { |
---|
310 | mat M; |
---|
311 | ldmat Lam; |
---|
312 | ldmat Vz; |
---|
313 | factorize ( M, Vz, Lam ); |
---|
314 | if( log_level[logvartheta] ) |
---|
315 | log_level.store( logvartheta, cvectorize ( est_theta_cov().to_mat() ) ); |
---|
316 | } |
---|
317 | } |
---|
318 | |
---|
319 | void egiw::from_setting ( const Setting &set ) { |
---|
320 | epdf::from_setting ( set ); |
---|
321 | UI::get ( dimx, set, "dimx", UI::compulsory ); |
---|
322 | if ( !UI::get ( nu, set, "nu", UI::optional ) ) { |
---|
323 | nu = -1; |
---|
324 | } |
---|
325 | mat Vful; |
---|
326 | if (set.exists("V.L")){ |
---|
327 | UI::get(V, set, "V", UI::optional); |
---|
328 | } else { |
---|
329 | if ( !UI::get ( Vful, set, "fV", UI::optional ) ) { |
---|
330 | vec dV; |
---|
331 | UI::get ( dV, set, "dV", UI::compulsory ); |
---|
332 | set_parameters ( dimx, ldmat ( dV ), nu ); |
---|
333 | |
---|
334 | } else { |
---|
335 | set_parameters ( dimx, Vful, nu ); |
---|
336 | } |
---|
337 | } |
---|
338 | } |
---|
339 | |
---|
340 | void egiw::to_setting ( Setting& set ) const { |
---|
341 | epdf::to_setting ( set ); |
---|
342 | UI::save ( dimx, set, "dimx" ); |
---|
343 | UI::save ( V, set, "V" ); |
---|
344 | UI::save ( nu, set, "nu" ); |
---|
345 | }; |
---|
346 | |
---|
347 | void egiw::validate() { |
---|
348 | eEF::validate(); |
---|
349 | nPsi = V.rows() - dimx; |
---|
350 | dim = dimx * ( dimx + nPsi ); |
---|
351 | if (dim<dimx) {bdm_error("Check if matrix V has correct dimension");} |
---|
352 | |
---|
353 | if ( nu < 0 ) { |
---|
354 | nu = 0.1 + nPsi + 2 * dimx + 2; // +2 assures finite expected value of R |
---|
355 | // terms before that are sufficient for finite normalization |
---|
356 | } |
---|
357 | |
---|
358 | // check sizes, rvs etc. |
---|
359 | // also check if RV are meaningful!!! |
---|
360 | // meaningful = rv for theta and rv for r are split! |
---|
361 | } |
---|
362 | void multiBM::bayes ( const vec &yt, const vec &cond ) { |
---|
363 | if ( frg < 1.0 ) { |
---|
364 | beta *= frg; |
---|
365 | last_lognc = est.lognc(); |
---|
366 | } |
---|
367 | beta += yt; |
---|
368 | if ( evalll ) { |
---|
369 | ll = est.lognc() - last_lognc; |
---|
370 | } |
---|
371 | } |
---|
372 | |
---|
373 | double multiBM::logpred ( const vec &yt ) const { |
---|
374 | eDirich pred ( est ); |
---|
375 | vec &beta = pred._beta(); |
---|
376 | |
---|
377 | double lll; |
---|
378 | if ( frg < 1.0 ) { |
---|
379 | beta *= frg; |
---|
380 | lll = pred.lognc(); |
---|
381 | } else if ( evalll ) { |
---|
382 | lll = last_lognc; |
---|
383 | } else { |
---|
384 | lll = pred.lognc(); |
---|
385 | } |
---|
386 | |
---|
387 | beta += yt; |
---|
388 | return pred.lognc() - lll; |
---|
389 | } |
---|
390 | void multiBM::flatten ( const BMEF* B, double weight=1.0 ) { |
---|
391 | const multiBM* E = dynamic_cast<const multiBM*> ( B ); |
---|
392 | // sum(beta) should be equal to sum(B.beta) |
---|
393 | const vec &Eb = E->beta;//const_cast<multiBM*> ( E )->_beta(); |
---|
394 | beta *= ( sum ( Eb ) / sum ( beta ) * weight); |
---|
395 | if ( evalll ) { |
---|
396 | last_lognc = est.lognc(); |
---|
397 | } |
---|
398 | } |
---|
399 | |
---|
400 | vec egamma::sample() const { |
---|
401 | vec smp ( dim ); |
---|
402 | int i; |
---|
403 | |
---|
404 | for ( i = 0; i < dim; i++ ) { |
---|
405 | if ( beta ( i ) > std::numeric_limits<double>::epsilon() ) { |
---|
406 | GamRNG.setup ( alpha ( i ), beta ( i ) ); |
---|
407 | } else { |
---|
408 | GamRNG.setup ( alpha ( i ), std::numeric_limits<double>::epsilon() ); |
---|
409 | } |
---|
410 | #pragma omp critical |
---|
411 | smp ( i ) = GamRNG(); |
---|
412 | } |
---|
413 | |
---|
414 | return smp; |
---|
415 | } |
---|
416 | |
---|
417 | |
---|
418 | double egamma::evallog_nn ( const vec &val ) const { |
---|
419 | double res = 0.0; //the rest will be added |
---|
420 | int i; |
---|
421 | |
---|
422 | if ( any ( val <= 0. ) ) return -inf; |
---|
423 | if ( any ( beta <= 0. ) ) return -inf; |
---|
424 | for ( i = 0; i < dim; i++ ) { |
---|
425 | res += ( alpha ( i ) - 1 ) * std::log ( val ( i ) ) - beta ( i ) * val ( i ); |
---|
426 | } |
---|
427 | return res; |
---|
428 | } |
---|
429 | |
---|
430 | double egamma::lognc() const { |
---|
431 | double res = 0.0; //will be added |
---|
432 | int i; |
---|
433 | |
---|
434 | for ( i = 0; i < dim; i++ ) { |
---|
435 | res += lgamma ( alpha ( i ) ) - alpha ( i ) * std::log ( beta ( i ) ) ; |
---|
436 | } |
---|
437 | |
---|
438 | return res; |
---|
439 | } |
---|
440 | |
---|
441 | void egamma::from_setting ( const Setting &set ) { |
---|
442 | epdf::from_setting ( set ); // reads rv |
---|
443 | UI::get ( alpha, set, "alpha", UI::compulsory ); |
---|
444 | UI::get ( beta, set, "beta", UI::compulsory ); |
---|
445 | } |
---|
446 | |
---|
447 | void egamma::to_setting ( Setting &set ) const |
---|
448 | { |
---|
449 | epdf::to_setting( set ); |
---|
450 | UI::save( alpha, set, "alpha" ); |
---|
451 | UI::save( beta, set, "beta" ); |
---|
452 | } |
---|
453 | |
---|
454 | |
---|
455 | void egamma::validate() { |
---|
456 | eEF::validate(); |
---|
457 | bdm_assert ( alpha.length() == beta.length(), "parameters do not match" ); |
---|
458 | dim = alpha.length(); |
---|
459 | } |
---|
460 | |
---|
461 | void mgamma::set_parameters ( double k0, const vec &beta0 ) { |
---|
462 | k = k0; |
---|
463 | iepdf.set_parameters ( k * ones ( beta0.length() ), beta0 ); |
---|
464 | } |
---|
465 | |
---|
466 | |
---|
467 | void mgamma::from_setting ( const Setting &set ) { |
---|
468 | pdf::from_setting ( set ); // reads rv and rvc |
---|
469 | vec betatmp; // ugly but necessary |
---|
470 | UI::get ( betatmp, set, "beta", UI::compulsory ); |
---|
471 | UI::get ( k, set, "k", UI::compulsory ); |
---|
472 | set_parameters ( k, betatmp ); |
---|
473 | } |
---|
474 | |
---|
475 | void mgamma::to_setting (Setting &set) const { |
---|
476 | pdf::to_setting(set); |
---|
477 | UI::save( _beta, set, "beta"); |
---|
478 | UI::save( k, set, "k"); |
---|
479 | |
---|
480 | } |
---|
481 | |
---|
482 | void mgamma::validate() { |
---|
483 | pdf_internal<egamma>::validate(); |
---|
484 | |
---|
485 | dim = _beta.length(); |
---|
486 | dimc = _beta.length(); |
---|
487 | } |
---|
488 | |
---|
489 | void eEmp::resample ( RESAMPLING_METHOD method ) { |
---|
490 | ivec ind = zeros_i ( n ); |
---|
491 | bdm::resample(w,ind,method); |
---|
492 | // copy the internals according to ind |
---|
493 | for (int i = 0; i < n; i++ ) { |
---|
494 | if ( ind ( i ) != i ) { |
---|
495 | samples ( i ) = samples ( ind ( i ) ); |
---|
496 | } |
---|
497 | w ( i ) = 1.0 / n; |
---|
498 | } |
---|
499 | } |
---|
500 | |
---|
501 | void resample ( const vec &w, ivec &ind, RESAMPLING_METHOD method ) { |
---|
502 | int n = w.length(); |
---|
503 | ind = zeros_i ( n ); |
---|
504 | ivec N_babies = zeros_i ( n ); |
---|
505 | vec cumDist = cumsum ( w ); |
---|
506 | vec u ( n ); |
---|
507 | int i, j, parent; |
---|
508 | double u0; |
---|
509 | |
---|
510 | switch ( method ) { |
---|
511 | case MULTINOMIAL: |
---|
512 | u ( n - 1 ) = pow ( UniRNG.sample(), 1.0 / n ); |
---|
513 | |
---|
514 | for ( i = n - 2; i >= 0; i-- ) { |
---|
515 | u ( i ) = u ( i + 1 ) * pow ( UniRNG.sample(), 1.0 / ( i + 1 ) ); |
---|
516 | } |
---|
517 | |
---|
518 | break; |
---|
519 | |
---|
520 | case STRATIFIED: |
---|
521 | |
---|
522 | for ( i = 0; i < n; i++ ) { |
---|
523 | u ( i ) = ( i + UniRNG.sample() ) / n; |
---|
524 | } |
---|
525 | |
---|
526 | break; |
---|
527 | |
---|
528 | case SYSTEMATIC: |
---|
529 | u0 = UniRNG.sample(); |
---|
530 | |
---|
531 | for ( i = 0; i < n; i++ ) { |
---|
532 | u ( i ) = ( i + u0 ) / n; |
---|
533 | } |
---|
534 | |
---|
535 | break; |
---|
536 | |
---|
537 | default: |
---|
538 | bdm_error ( "PF::resample(): Unknown resampling method" ); |
---|
539 | } |
---|
540 | |
---|
541 | // U is now full |
---|
542 | j = 0; |
---|
543 | |
---|
544 | for ( i = 0; i < n; i++ ) { |
---|
545 | while ( u ( i ) > cumDist ( j ) ) j++; |
---|
546 | |
---|
547 | N_babies ( j ) ++; |
---|
548 | } |
---|
549 | // We have assigned new babies for each Particle |
---|
550 | // Now, we fill the resulting index such that: |
---|
551 | // * particles with at least one baby should not move * |
---|
552 | // This assures that reassignment can be done inplace; |
---|
553 | |
---|
554 | // find the first parent; |
---|
555 | parent = 0; |
---|
556 | while ( N_babies ( parent ) == 0 ) parent++; |
---|
557 | |
---|
558 | // Build index |
---|
559 | for ( i = 0; i < n; i++ ) { |
---|
560 | if ( N_babies ( i ) > 0 ) { |
---|
561 | ind ( i ) = i; |
---|
562 | N_babies ( i ) --; //this index was now replicated; |
---|
563 | } else { |
---|
564 | // test if the parent has been fully replicated |
---|
565 | // if yes, find the next one |
---|
566 | while ( ( N_babies ( parent ) == 0 ) || ( N_babies ( parent ) == 1 && parent > i ) ) parent++; |
---|
567 | |
---|
568 | // Replicate parent |
---|
569 | ind ( i ) = parent; |
---|
570 | |
---|
571 | N_babies ( parent ) --; //this index was now replicated; |
---|
572 | } |
---|
573 | |
---|
574 | } |
---|
575 | } |
---|
576 | |
---|
577 | void eEmp::set_statistics ( const vec &w0, const epdf &epdf0 ) { |
---|
578 | dim = epdf0.dimension(); |
---|
579 | w = w0; |
---|
580 | w /= sum ( w0 );//renormalize |
---|
581 | n = w.length(); |
---|
582 | samples.set_size ( n ); |
---|
583 | |
---|
584 | for ( int i = 0; i < n; i++ ) { |
---|
585 | samples ( i ) = epdf0.sample(); |
---|
586 | } |
---|
587 | } |
---|
588 | |
---|
589 | void eEmp::set_samples ( const epdf* epdf0 ) { |
---|
590 | w = 1; |
---|
591 | w /= sum ( w );//renormalize |
---|
592 | |
---|
593 | for ( int i = 0; i < n; i++ ) { |
---|
594 | samples ( i ) = epdf0->sample(); |
---|
595 | } |
---|
596 | } |
---|
597 | |
---|
598 | void migamma_ref::from_setting ( const Setting &set ) { |
---|
599 | migamma::from_setting(set); |
---|
600 | vec ref; |
---|
601 | double k,l; |
---|
602 | |
---|
603 | UI::get ( ref, set, "ref" , UI::compulsory ); |
---|
604 | UI::get( k, set, "k", UI::compulsory ); |
---|
605 | UI::get( l, set, "l", UI::compulsory ); |
---|
606 | set_parameters ( k, ref, l ); |
---|
607 | } |
---|
608 | |
---|
609 | void migamma_ref::to_setting (Setting &set) const { |
---|
610 | migamma::to_setting(set); |
---|
611 | UI::save ( pow ( refl, 1/(1.0 - l) ), set, "ref"); |
---|
612 | UI::save(l,set,"l"); |
---|
613 | UI::save(k,set,"k"); |
---|
614 | } |
---|
615 | |
---|
616 | void mlognorm::from_setting ( const Setting &set ) { |
---|
617 | pdf_internal<elognorm>::from_setting(set); |
---|
618 | vec mu0; |
---|
619 | double k; |
---|
620 | UI::get ( mu0, set, "mu0", UI::compulsory ); |
---|
621 | UI::get( k, set, "k", UI::compulsory ); |
---|
622 | set_parameters ( mu0.length(), k ); |
---|
623 | condition ( mu0 ); |
---|
624 | } |
---|
625 | |
---|
626 | void mlognorm::to_setting (Setting &set) const { |
---|
627 | pdf_internal<elognorm>::to_setting(set); |
---|
628 | UI::save ( exp(mu + sig2), set, "mu0"); |
---|
629 | |
---|
630 | // inversion of sig2 = 0.5 * log ( k * k + 1 ); |
---|
631 | double k = sqrt( exp( 2 * sig2 ) - 1 ); |
---|
632 | UI::save(k,set,"k"); |
---|
633 | } |
---|
634 | |
---|
635 | |
---|
636 | void mlstudent::condition ( const vec &cond ) { |
---|
637 | if ( cond.length() > 0 ) { |
---|
638 | iepdf._mu() = A * cond + mu_const; |
---|
639 | } else { |
---|
640 | iepdf._mu() = mu_const; |
---|
641 | } |
---|
642 | double zeta; |
---|
643 | //ugly hack! |
---|
644 | if ( ( cond.length() + 1 ) == Lambda.rows() ) { |
---|
645 | zeta = Lambda.invqform ( concat ( cond, vec_1 ( 1.0 ) ) ); |
---|
646 | } else { |
---|
647 | zeta = Lambda.invqform ( cond ); |
---|
648 | } |
---|
649 | _R = Re; |
---|
650 | _R *= ( 1 + zeta );// / ( nu ); << nu is in Re!!!!!! |
---|
651 | } |
---|
652 | |
---|
653 | void eEmp::qbounds ( vec &lb, vec &ub, double perc ) const { |
---|
654 | // lb in inf so than it will be pushed below; |
---|
655 | lb.set_size ( dim ); |
---|
656 | ub.set_size ( dim ); |
---|
657 | lb = std::numeric_limits<double>::infinity(); |
---|
658 | ub = -std::numeric_limits<double>::infinity(); |
---|
659 | int j; |
---|
660 | for ( int i = 0; i < n; i++ ) { |
---|
661 | for ( j = 0; j < dim; j++ ) { |
---|
662 | if ( samples ( i ) ( j ) < lb ( j ) ) { |
---|
663 | lb ( j ) = samples ( i ) ( j ); |
---|
664 | } |
---|
665 | if ( samples ( i ) ( j ) > ub ( j ) ) { |
---|
666 | ub ( j ) = samples ( i ) ( j ); |
---|
667 | } |
---|
668 | } |
---|
669 | } |
---|
670 | } |
---|
671 | |
---|
672 | void eEmp::to_setting ( Setting &set ) const { |
---|
673 | epdf::to_setting( set ); |
---|
674 | UI::save ( samples, set, "samples" ); |
---|
675 | UI::save ( w, set, "w" ); |
---|
676 | } |
---|
677 | |
---|
678 | void eEmp::from_setting ( const Setting &set ) { |
---|
679 | epdf::from_setting( set ); |
---|
680 | |
---|
681 | UI::get( samples, set, "samples", UI::compulsory ); |
---|
682 | UI::get ( w, set, "w", UI::compulsory ); |
---|
683 | } |
---|
684 | |
---|
685 | void eEmp::validate () { |
---|
686 | epdf::validate(); |
---|
687 | bdm_assert (samples.length()==w.length(),"samples and weigths are of different lengths"); |
---|
688 | n = w.length(); |
---|
689 | if (n>0) |
---|
690 | pdf::dim = samples ( 0 ).length(); |
---|
691 | } |
---|
692 | |
---|
693 | void eDirich::from_setting ( const Setting &set ) { |
---|
694 | epdf::from_setting ( set ); |
---|
695 | UI::get ( beta, set, "beta", UI::compulsory ); |
---|
696 | } |
---|
697 | |
---|
698 | void eDirich::validate() { |
---|
699 | //check rv |
---|
700 | eEF::validate(); |
---|
701 | dim = beta.length(); |
---|
702 | } |
---|
703 | |
---|
704 | void eDirich::to_setting ( Setting &set ) const |
---|
705 | { |
---|
706 | eEF::to_setting( set ); |
---|
707 | UI::save( beta, set, "beta" ); |
---|
708 | } |
---|
709 | |
---|
710 | void euni::from_setting ( const Setting &set ) { |
---|
711 | epdf::from_setting ( set ); // reads rv and rvc |
---|
712 | |
---|
713 | UI::get ( high, set, "high", UI::compulsory ); |
---|
714 | UI::get ( low, set, "low", UI::compulsory ); |
---|
715 | set_parameters ( low, high ); |
---|
716 | |
---|
717 | } |
---|
718 | |
---|
719 | void euni::to_setting (Setting &set) const { |
---|
720 | epdf::to_setting ( set ); |
---|
721 | UI::save ( high, set, "high" ); |
---|
722 | UI::save ( low, set, "low" ); |
---|
723 | } |
---|
724 | |
---|
725 | void euni::validate() { |
---|
726 | epdf::validate(); |
---|
727 | bdm_assert ( high.length() == low.length(), "Incompatible high and low vectors" ); |
---|
728 | dim = high.length(); |
---|
729 | bdm_assert ( min ( distance ) > 0.0, "bad support" ); |
---|
730 | } |
---|
731 | |
---|
732 | void mgdirac::from_setting(const Setting& set) { |
---|
733 | pdf::from_setting(set); |
---|
734 | g=UI::build<fnc>(set,"g",UI::compulsory); |
---|
735 | } |
---|
736 | |
---|
737 | void mgdirac::to_setting(Setting &set) const { |
---|
738 | pdf::to_setting(set); |
---|
739 | UI::save(g.get(), set, "g"); |
---|
740 | } |
---|
741 | |
---|
742 | void mgdirac::validate() { |
---|
743 | pdf::validate(); |
---|
744 | dim = g->dimension(); |
---|
745 | dimc = g->dimensionc(); |
---|
746 | } |
---|
747 | |
---|
748 | void mDirich::from_setting ( const Setting &set ) { |
---|
749 | pdf::from_setting ( set ); // reads rv and rvc |
---|
750 | if ( _rv()._dsize() > 0 ) { |
---|
751 | rvc = _rv().copy_t ( -1 ); |
---|
752 | } |
---|
753 | vec beta0; |
---|
754 | if ( !UI::get ( beta0, set, "beta0", UI::optional ) ) { |
---|
755 | beta0 = ones ( _rv()._dsize() ); |
---|
756 | } |
---|
757 | if ( !UI::get ( betac, set, "betac", UI::optional ) ) { |
---|
758 | betac = 0.1 * ones ( _rv()._dsize() ); |
---|
759 | } |
---|
760 | _beta = beta0; |
---|
761 | |
---|
762 | UI::get ( k, set, "k", UI::compulsory ); |
---|
763 | } |
---|
764 | |
---|
765 | void mDirich::to_setting (Setting &set) const { |
---|
766 | pdf::to_setting(set); |
---|
767 | UI::save( _beta, set, "beta0"); |
---|
768 | UI::save( betac, set, "betac"); |
---|
769 | UI::save ( k, set, "k" ); |
---|
770 | } |
---|
771 | |
---|
772 | |
---|
773 | void mDirich::validate() { |
---|
774 | pdf_internal<eDirich>::validate(); |
---|
775 | bdm_assert ( _beta.length() == betac.length(), "beta0 and betac are not compatible" ); |
---|
776 | if ( _rv()._dsize() > 0 ) { |
---|
777 | bdm_assert ( ( _rv()._dsize() == dimension() ) , "Size of rv does not match with beta" ); |
---|
778 | } |
---|
779 | dimc = _beta.length(); |
---|
780 | } |
---|
781 | |
---|
782 | |
---|
783 | void mBeta::from_setting ( const Setting &set ) { |
---|
784 | pdf::from_setting ( set ); // reads rv and rvc |
---|
785 | if ( _rv()._dsize() > 0 ) { |
---|
786 | rvc = _rv().copy_t ( -1 ); |
---|
787 | } |
---|
788 | if ( !UI::get ( iepdf.beta, set, "beta", UI::optional ) ) { |
---|
789 | iepdf.beta = ones ( _rv()._dsize() ); |
---|
790 | } |
---|
791 | if ( !UI::get ( iepdf.alpha, set, "alpha", UI::optional ) ) { |
---|
792 | iepdf.alpha = ones ( _rv()._dsize() ); |
---|
793 | } |
---|
794 | if ( !UI::get ( betac, set, "betac", UI::optional ) ) { |
---|
795 | betac = 0.1 * ones ( _rv()._dsize() ); |
---|
796 | } |
---|
797 | |
---|
798 | UI::get ( k, set, "k", UI::compulsory ); |
---|
799 | } |
---|
800 | |
---|
801 | void mBeta::to_setting (Setting &set) const { |
---|
802 | pdf::to_setting(set); |
---|
803 | UI::save( iepdf.beta, set, "beta"); |
---|
804 | UI::save( betac, set, "betac"); |
---|
805 | UI::save ( k, set, "k" ); |
---|
806 | } |
---|
807 | |
---|
808 | } |
---|