1 | // |
---|
2 | // C++ Implementation: itpp_ext |
---|
3 | // |
---|
4 | // Description: |
---|
5 | // |
---|
6 | // |
---|
7 | // Author: smidl <smidl@utia.cas.cz>, (C) 2008 |
---|
8 | // |
---|
9 | // Copyright: See COPYING file that comes with this distribution |
---|
10 | // |
---|
11 | // |
---|
12 | |
---|
13 | #include "itpp_ext.h" |
---|
14 | |
---|
15 | #ifndef M_PI |
---|
16 | #define M_PI 3.14159265358979323846 |
---|
17 | #endif |
---|
18 | // from algebra/lapack.h |
---|
19 | |
---|
20 | extern "C" { /* QR factorization of a general matrix A */ |
---|
21 | void dgeqrf_ ( int *m, int *n, double *a, int *lda, double *tau, double *work, |
---|
22 | int *lwork, int *info ); |
---|
23 | }; |
---|
24 | |
---|
25 | namespace itpp { |
---|
26 | Array<int> to_Arr ( const ivec &indices ) { |
---|
27 | Array<int> a ( indices.size() ); |
---|
28 | for ( int i = 0; i < a.size(); i++ ) { |
---|
29 | a ( i ) = indices ( i ); |
---|
30 | } |
---|
31 | return a; |
---|
32 | } |
---|
33 | |
---|
34 | ivec linspace ( int from, int to ) { |
---|
35 | int n=to-from+1; |
---|
36 | int i; |
---|
37 | it_assert_debug ( n>0,"wrong linspace" ); |
---|
38 | ivec iv ( n ); for ( i=0;i<n;i++ ) iv ( i ) =from+i; |
---|
39 | return iv; |
---|
40 | }; |
---|
41 | |
---|
42 | void set_subvector ( vec &ov, const ivec &iv, const vec &v ) |
---|
43 | { |
---|
44 | it_assert_debug ( ( iv.length() <=v.length() ), |
---|
45 | "Vec<>::set_subvector(ivec, vec<Num_T>): Indexing out " |
---|
46 | "of range of v" ); |
---|
47 | for ( int i = 0; i < iv.length(); i++ ) { |
---|
48 | it_assert_debug ( iv ( i ) <ov.length(), |
---|
49 | "Vec<>::set_subvector(ivec, vec<Num_T>): Indexing out " |
---|
50 | "of range of v" ); |
---|
51 | ov ( iv ( i ) ) = v ( i ); |
---|
52 | } |
---|
53 | } |
---|
54 | |
---|
55 | vec get_vec(const vec &v, const ivec &indexlist){ |
---|
56 | int size = indexlist.size(); |
---|
57 | vec temp(size); |
---|
58 | for (int i = 0; i < size; ++i) { |
---|
59 | temp(i) = v._data()[indexlist(i)]; |
---|
60 | } |
---|
61 | return temp; |
---|
62 | } |
---|
63 | |
---|
64 | // Gamma |
---|
65 | #define log std::log |
---|
66 | #define exp std::exp |
---|
67 | #define sqrt std::sqrt |
---|
68 | #define R_FINITE std::isfinite |
---|
69 | |
---|
70 | bvec operator>(const vec &t1, const vec &t2) { |
---|
71 | it_assert_debug(t1.length() == t2.length(), "Vec<>::operator>(): different size of vectors"); |
---|
72 | bvec temp(t1.length()); |
---|
73 | for (int i = 0; i < t1.length(); i++) |
---|
74 | temp(i) = (t1[i] > t2[i]); |
---|
75 | return temp; |
---|
76 | } |
---|
77 | |
---|
78 | bvec operator<(const vec &t1, const vec &t2) { |
---|
79 | it_assert_debug(t1.length() == t2.length(), "Vec<>::operator>(): different size of vectors"); |
---|
80 | bvec temp(t1.length()); |
---|
81 | for (int i = 0; i < t1.length(); i++) |
---|
82 | temp(i) = (t1[i] < t2[i]); |
---|
83 | return temp; |
---|
84 | } |
---|
85 | |
---|
86 | |
---|
87 | bvec operator& ( const bvec &a, const bvec &b ) { |
---|
88 | it_assert_debug ( b.size() ==a.size(), "operator&(): Vectors of different lengths" ); |
---|
89 | |
---|
90 | bvec temp ( a.size() ); |
---|
91 | for ( int i = 0;i < a.size();i++ ) { |
---|
92 | temp ( i ) = a ( i ) & b ( i ); |
---|
93 | } |
---|
94 | return temp; |
---|
95 | } |
---|
96 | |
---|
97 | bvec operator| ( const bvec &a, const bvec &b ) { |
---|
98 | it_assert_debug ( b.size() !=a.size(), "operator&(): Vectors of different lengths" ); |
---|
99 | |
---|
100 | bvec temp ( a.size() ); |
---|
101 | for ( int i = 0;i < a.size();i++ ) { |
---|
102 | temp ( i ) = a ( i ) | b ( i ); |
---|
103 | } |
---|
104 | return temp; |
---|
105 | } |
---|
106 | |
---|
107 | //#if 0 |
---|
108 | Gamma_RNG::Gamma_RNG ( double a, double b ) { |
---|
109 | setup ( a,b ); |
---|
110 | } |
---|
111 | double Gamma_RNG::sample() { |
---|
112 | //A copy of rgamma code from the R package!! |
---|
113 | // |
---|
114 | |
---|
115 | /* Constants : */ |
---|
116 | const static double sqrt32 = 5.656854; |
---|
117 | const static double exp_m1 = 0.36787944117144232159;/* exp(-1) = 1/e */ |
---|
118 | |
---|
119 | /* Coefficients q[k] - for q0 = sum(q[k]*a^(-k)) |
---|
120 | * Coefficients a[k] - for q = q0+(t*t/2)*sum(a[k]*v^k) |
---|
121 | * Coefficients e[k] - for exp(q)-1 = sum(e[k]*q^k) |
---|
122 | */ |
---|
123 | const static double q1 = 0.04166669; |
---|
124 | const static double q2 = 0.02083148; |
---|
125 | const static double q3 = 0.00801191; |
---|
126 | const static double q4 = 0.00144121; |
---|
127 | const static double q5 = -7.388e-5; |
---|
128 | const static double q6 = 2.4511e-4; |
---|
129 | const static double q7 = 2.424e-4; |
---|
130 | |
---|
131 | const static double a1 = 0.3333333; |
---|
132 | const static double a2 = -0.250003; |
---|
133 | const static double a3 = 0.2000062; |
---|
134 | const static double a4 = -0.1662921; |
---|
135 | const static double a5 = 0.1423657; |
---|
136 | const static double a6 = -0.1367177; |
---|
137 | const static double a7 = 0.1233795; |
---|
138 | |
---|
139 | /* State variables [FIXME for threading!] :*/ |
---|
140 | static double aa = 0.; |
---|
141 | static double aaa = 0.; |
---|
142 | static double s, s2, d; /* no. 1 (step 1) */ |
---|
143 | static double q0, b, si, c;/* no. 2 (step 4) */ |
---|
144 | |
---|
145 | double e, p, q, r, t, u, v, w, x, ret_val; |
---|
146 | double a=alpha; |
---|
147 | double scale=1.0/beta; |
---|
148 | |
---|
149 | if ( !R_FINITE ( a ) || !R_FINITE ( scale ) || a < 0.0 || scale <= 0.0 ) |
---|
150 | {it_error ( "Gamma_RNG wrong parameters" );} |
---|
151 | |
---|
152 | if ( a < 1. ) { /* GS algorithm for parameters a < 1 */ |
---|
153 | if ( a == 0 ) |
---|
154 | return 0.; |
---|
155 | e = 1.0 + exp_m1 * a; |
---|
156 | for ( ;; ) { //VS repeat |
---|
157 | p = e * unif_rand(); |
---|
158 | if ( p >= 1.0 ) { |
---|
159 | x = -log ( ( e - p ) / a ); |
---|
160 | if ( exp_rand() >= ( 1.0 - a ) * log ( x ) ) |
---|
161 | break; |
---|
162 | } |
---|
163 | else { |
---|
164 | x = exp ( log ( p ) / a ); |
---|
165 | if ( exp_rand() >= x ) |
---|
166 | break; |
---|
167 | } |
---|
168 | } |
---|
169 | return scale * x; |
---|
170 | } |
---|
171 | |
---|
172 | /* --- a >= 1 : GD algorithm --- */ |
---|
173 | |
---|
174 | /* Step 1: Recalculations of s2, s, d if a has changed */ |
---|
175 | if ( a != aa ) { |
---|
176 | aa = a; |
---|
177 | s2 = a - 0.5; |
---|
178 | s = sqrt ( s2 ); |
---|
179 | d = sqrt32 - s * 12.0; |
---|
180 | } |
---|
181 | /* Step 2: t = standard normal deviate, |
---|
182 | x = (s,1/2) -normal deviate. */ |
---|
183 | |
---|
184 | /* immediate acceptance (i) */ |
---|
185 | t = norm_rand(); |
---|
186 | x = s + 0.5 * t; |
---|
187 | ret_val = x * x; |
---|
188 | if ( t >= 0.0 ) |
---|
189 | return scale * ret_val; |
---|
190 | |
---|
191 | /* Step 3: u = 0,1 - uniform sample. squeeze acceptance (s) */ |
---|
192 | u = unif_rand(); |
---|
193 | if ( ( d * u ) <= ( t * t * t ) ) |
---|
194 | return scale * ret_val; |
---|
195 | |
---|
196 | /* Step 4: recalculations of q0, b, si, c if necessary */ |
---|
197 | |
---|
198 | if ( a != aaa ) { |
---|
199 | aaa = a; |
---|
200 | r = 1.0 / a; |
---|
201 | q0 = ( ( ( ( ( ( q7 * r + q6 ) * r + q5 ) * r + q4 ) * r + q3 ) * r |
---|
202 | + q2 ) * r + q1 ) * r; |
---|
203 | |
---|
204 | /* Approximation depending on size of parameter a */ |
---|
205 | /* The constants in the expressions for b, si and c */ |
---|
206 | /* were established by numerical experiments */ |
---|
207 | |
---|
208 | if ( a <= 3.686 ) { |
---|
209 | b = 0.463 + s + 0.178 * s2; |
---|
210 | si = 1.235; |
---|
211 | c = 0.195 / s - 0.079 + 0.16 * s; |
---|
212 | } |
---|
213 | else if ( a <= 13.022 ) { |
---|
214 | b = 1.654 + 0.0076 * s2; |
---|
215 | si = 1.68 / s + 0.275; |
---|
216 | c = 0.062 / s + 0.024; |
---|
217 | } |
---|
218 | else { |
---|
219 | b = 1.77; |
---|
220 | si = 0.75; |
---|
221 | c = 0.1515 / s; |
---|
222 | } |
---|
223 | } |
---|
224 | /* Step 5: no quotient test if x not positive */ |
---|
225 | |
---|
226 | if ( x > 0.0 ) { |
---|
227 | /* Step 6: calculation of v and quotient q */ |
---|
228 | v = t / ( s + s ); |
---|
229 | if ( fabs ( v ) <= 0.25 ) |
---|
230 | q = q0 + 0.5 * t * t * ( ( ( ( ( ( a7 * v + a6 ) * v + a5 ) * v + a4 ) * v |
---|
231 | + a3 ) * v + a2 ) * v + a1 ) * v; |
---|
232 | else |
---|
233 | q = q0 - s * t + 0.25 * t * t + ( s2 + s2 ) * log ( 1.0 + v ); |
---|
234 | |
---|
235 | |
---|
236 | /* Step 7: quotient acceptance (q) */ |
---|
237 | if ( log ( 1.0 - u ) <= q ) |
---|
238 | return scale * ret_val; |
---|
239 | } |
---|
240 | |
---|
241 | for ( ;; ) { //VS repeat |
---|
242 | /* Step 8: e = standard exponential deviate |
---|
243 | * u = 0,1 -uniform deviate |
---|
244 | * t = (b,si)-double exponential (laplace) sample */ |
---|
245 | e = exp_rand(); |
---|
246 | u = unif_rand(); |
---|
247 | u = u + u - 1.0; |
---|
248 | if ( u < 0.0 ) |
---|
249 | t = b - si * e; |
---|
250 | else |
---|
251 | t = b + si * e; |
---|
252 | /* Step 9: rejection if t < tau(1) = -0.71874483771719 */ |
---|
253 | if ( t >= -0.71874483771719 ) { |
---|
254 | /* Step 10: calculation of v and quotient q */ |
---|
255 | v = t / ( s + s ); |
---|
256 | if ( fabs ( v ) <= 0.25 ) |
---|
257 | q = q0 + 0.5 * t * t * |
---|
258 | ( ( ( ( ( ( a7 * v + a6 ) * v + a5 ) * v + a4 ) * v + a3 ) * v |
---|
259 | + a2 ) * v + a1 ) * v; |
---|
260 | else |
---|
261 | q = q0 - s * t + 0.25 * t * t + ( s2 + s2 ) * log ( 1.0 + v ); |
---|
262 | /* Step 11: hat acceptance (h) */ |
---|
263 | /* (if q not positive go to step 8) */ |
---|
264 | if ( q > 0.0 ) { |
---|
265 | // TODO: w = expm1(q); |
---|
266 | w = exp ( q )-1; |
---|
267 | /* ^^^^^ original code had approximation with rel.err < 2e-7 */ |
---|
268 | /* if t is rejected sample again at step 8 */ |
---|
269 | if ( ( c * fabs ( u ) ) <= ( w * exp ( e - 0.5 * t * t ) ) ) |
---|
270 | break; |
---|
271 | } |
---|
272 | } |
---|
273 | } /* repeat .. until `t' is accepted */ |
---|
274 | x = s + 0.5 * t; |
---|
275 | return scale * x * x; |
---|
276 | } |
---|
277 | |
---|
278 | |
---|
279 | bool qr ( const mat &A, mat &R ) { |
---|
280 | int info; |
---|
281 | int m = A.rows(); |
---|
282 | int n = A.cols(); |
---|
283 | int lwork = n; |
---|
284 | int k = std::min ( m, n ); |
---|
285 | vec tau ( k ); |
---|
286 | vec work ( lwork ); |
---|
287 | |
---|
288 | R = A; |
---|
289 | |
---|
290 | // perform workspace query for optimum lwork value |
---|
291 | int lwork_tmp = -1; |
---|
292 | dgeqrf_ ( &m, &n, R._data(), &m, tau._data(), work._data(), &lwork_tmp, |
---|
293 | &info ); |
---|
294 | if ( info == 0 ) { |
---|
295 | lwork = static_cast<int> ( work ( 0 ) ); |
---|
296 | work.set_size ( lwork, false ); |
---|
297 | } |
---|
298 | dgeqrf_ ( &m, &n, R._data(), &m, tau._data(), work._data(), &lwork, &info ); |
---|
299 | |
---|
300 | // construct R |
---|
301 | for ( int i=0; i<m; i++ ) |
---|
302 | for ( int j=0; j<std::min ( i,n ); j++ ) |
---|
303 | R ( i,j ) = 0; |
---|
304 | |
---|
305 | return ( info==0 ); |
---|
306 | } |
---|
307 | |
---|
308 | //#endif |
---|
309 | std::string num2str(double d){ |
---|
310 | char tmp[20];//that should do |
---|
311 | sprintf(tmp,"%f",d); |
---|
312 | return std::string(tmp); |
---|
313 | }; |
---|
314 | std::string num2str(int i){ |
---|
315 | char tmp[10];//that should do |
---|
316 | sprintf(tmp,"%d",i); |
---|
317 | return std::string(tmp); |
---|
318 | }; |
---|
319 | |
---|
320 | // digamma |
---|
321 | // copied from C. Bonds' source |
---|
322 | #include <math.h> |
---|
323 | #define el 0.5772156649015329 |
---|
324 | |
---|
325 | double psi(double x) { |
---|
326 | double s,ps,xa,x2; |
---|
327 | int n,k; |
---|
328 | static double a[] = { |
---|
329 | -0.8333333333333e-01, |
---|
330 | 0.83333333333333333e-02, |
---|
331 | -0.39682539682539683e-02, |
---|
332 | 0.41666666666666667e-02, |
---|
333 | -0.75757575757575758e-02, |
---|
334 | 0.21092796092796093e-01, |
---|
335 | -0.83333333333333333e-01, |
---|
336 | 0.4432598039215686}; |
---|
337 | |
---|
338 | xa = fabs(x); |
---|
339 | s = 0.0; |
---|
340 | if ((x == (int)x) && (x <= 0.0)) { |
---|
341 | ps = 1e308; |
---|
342 | return ps; |
---|
343 | } |
---|
344 | if (xa == (int)xa) { |
---|
345 | n = xa; |
---|
346 | for (k=1;k<n;k++) { |
---|
347 | s += 1.0/k; |
---|
348 | } |
---|
349 | ps = s-el; |
---|
350 | } |
---|
351 | else if ((xa+0.5) == ((int)(xa+0.5))) { |
---|
352 | n = xa-0.5; |
---|
353 | for (k=1;k<=n;k++) { |
---|
354 | s += 1.0/(2.0*k-1.0); |
---|
355 | } |
---|
356 | ps = 2.0*s-el-1.386294361119891; |
---|
357 | } |
---|
358 | else { |
---|
359 | if (xa < 10.0) { |
---|
360 | n = 10-(int)xa; |
---|
361 | for (k=0;k<n;k++) { |
---|
362 | s += 1.0/(xa+k); |
---|
363 | } |
---|
364 | xa += n; |
---|
365 | } |
---|
366 | x2 = 1.0/(xa*xa); |
---|
367 | ps = log(xa)-0.5/xa+x2*(((((((a[7]*x2+a[6])*x2+a[5])*x2+ |
---|
368 | a[4])*x2+a[3])*x2+a[2])*x2+a[1])*x2+a[0]); |
---|
369 | ps -= s; |
---|
370 | } |
---|
371 | if (x < 0.0) |
---|
372 | ps = ps - M_PI*std::cos(M_PI*x)/std::sin(M_PI*x)-1.0/x; |
---|
373 | return ps; |
---|
374 | } |
---|
375 | |
---|
376 | } |
---|