-
-
Notifications
You must be signed in to change notification settings - Fork 59
/
Copy pathoptimization.cpp
1318 lines (1184 loc) · 35.2 KB
/
optimization.cpp
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
//
// C++ Implementation: optimization
//
// Description:
//
//
// Author: BUI Quang Minh, Steffen Klaere, Arndt von Haeseler <minh.bui@univie.ac.at>, (C) 2008
//
// Copyright: See COPYING file that comes with this distribution
//
//
#include "optimization.h"
#include <cmath>
#include <stdio.h>
#include <stdlib.h>
#include <iostream>
#include "lbfgsb/lbfgsb_new.h"
#include "tools.h"
using namespace std;
const double ERROR_X = 1.0e-4;
double ran1(long *idum);
double *new_vector(long nl, long nh);
void free_vector(double *v, long nl, long nh);
double **new_matrix(long nrl, long nrh, long ncl, long nch);
void free_matrix(double **m, long nrl, long nrh, long ncl, long nch);
void fixBound(double x[], double lower[], double upper[], int n);
#define NR_END 1
#define FREE_ARG char*
#define GET_PSUM \
for (n=1;n<=ndim;n++) {\
for (sum=0.0,m=1;m<=mpts;m++) sum += p[m][n];\
psum[n]=sum;}
/*
#define IA 16807
#define IM 2147483647
#define AM (1.0/IM)
#define IQ 127773
#define IR 2836
#define NTAB 32
#define NDIV (1+(IM-1)/NTAB)
#define EPS 1.2e-7
#define RNMX (1.0-EPS)
double ran1(long *idum) {
int j;
long k;
static long iy=0;
static long iv[NTAB];
double temp;
if (*idum <= 0 || !iy) {
if (-(*idum) < 1) *idum=1;
else *idum = -(*idum);
for (j=NTAB+7;j>=0;j--) {
k=(*idum)/IQ;
*idum=IA*(*idum-k*IQ)-IR*k;
if (*idum < 0) *idum += IM;
if (j < NTAB) iv[j] = *idum;
}
iy=iv[0];
}
k=(*idum)/IQ;
*idum=IA*(*idum-k*IQ)-IR*k;
if (*idum < 0) *idum += IM;
j=iy/NDIV;
iy=iv[j];
iv[j] = *idum;
if ((temp=AM*iy) > RNMX) return RNMX;
else return temp;
}
#undef IA
#undef IM
#undef AM
#undef IQ
#undef IR
#undef NTAB
#undef NDIV
#undef EPS
#undef RNMX
*/
long idum = 123456;
double tt;
void nrerror(const char *error_text)
/* Numerical Recipes standard error handler */
{
cerr << "NUMERICAL ERROR: " << error_text << endl;
//exit(1);
abort();
}
double *new_vector(long nl, long nh)
/* allocate a double vector with subscript range v[nl..nh] */
{
double *v;
v=(double *)malloc((size_t) ((nh-nl+1+NR_END)*sizeof(double)));
if (!v) nrerror("allocation failure in vector()");
return v-nl+NR_END;
}
double **new_matrix(long nrl, long nrh, long ncl, long nch)
/* allocate a double matrix with subscript range m[nrl..nrh][ncl..nch] */
{
long i, nrow=nrh-nrl+1,ncol=nch-ncl+1;
double **m;
/* allocate pointers to rows */
m=(double **) malloc((size_t)((nrow+NR_END)*sizeof(double*)));
if (!m) nrerror("allocation failure 1 in matrix()");
m += NR_END;
m -= nrl;
/* allocate rows and set pointers to them */
m[nrl]=(double *) malloc((size_t)((nrow*ncol+NR_END)*sizeof(double)));
if (!m[nrl]) nrerror("allocation failure 2 in matrix()");
m[nrl] += NR_END;
m[nrl] -= ncl;
for(i=nrl+1;i<=nrh;i++) m[i]=m[i-1]+ncol;
/* return pointer to array of pointers to rows */
return m;
}
void free_vector(double *v, long nl, long nh)
/* free a double vector allocated with vector() */
{
free((FREE_ARG) (v+nl-NR_END));
}
void free_matrix(double **m, long nrl, long nrh, long ncl, long nch)
/* free a double matrix allocated by dmatrix() */
{
free((FREE_ARG) (m[nrl]+ncl-NR_END));
free((FREE_ARG) (m+nrl-NR_END));
}
void fixBound(double x[], double lower[], double upper[], int n) {
for (int i = 1; i <= n; i++) {
if (x[i] < lower[i])
x[i] = lower[i];
else if (x[i] > upper[i])
x[i] = upper[i];
}
}
/**********************************************
Optimization routines
**********************************************/
Optimization::Optimization()
{
}
Optimization::~Optimization()
{
}
/*****************************************************
One dimensional optimization with Brent method
*****************************************************/
#define ITMAX 100
#define CGOLD 0.3819660
#define GOLD 1.618034
#define GLIMIT 100.0
#define TINY 1.0e-20
#define ZEPS 1.0e-10
#define SHFT(a,b,c,d) (a)=(b);(b)=(c);(c)=(d);
#define SIGN(a,b) ((b) >= 0.0 ? fabs(a) : -fabs(a))
#define TRY_AITKEN 0
/* Brents method in one dimension */
double Optimization::brent_opt (double ax, double bx, double cx, double tol,
double *foptx, double *f2optx, double fax, double fbx, double fcx) {
int iter;
double a,b,d=0,etemp,fu,fv,fw,fx,p,q,r,tol1,tol2,u,v,w,x,xm;
double xw,wv,vx;
double e=0.0;
a=(ax < cx ? ax : cx);
b=(ax > cx ? ax : cx);
x=bx;
fx=fbx;
if (fax < fcx) {
w=ax;
fw=fax;
v=cx;
fv=fcx;
} else {
w=cx;
fw=fcx;
v=ax;
fv=fax;
}
#if TRY_AITKEN
double delta = 0; //change in U
double lastDelta = 0; //previous change in U
double lastU = 0; //previous value of U
#endif
for (iter=1;iter<=ITMAX;iter++) {
xm=0.5*(a+b);
tol2=2.0*(tol1=tol*fabs(x)+ZEPS);
if (fabs(x-xm) <= (tol2-0.5*(b-a))) {
*foptx = fx;
xw = x-w;
wv = w-v;
vx = v-x;
*f2optx = 2.0*(fv*xw + fx*wv + fw*vx)/
(v*v*xw + x*x*wv + w*w*vx);
return x;
}
if (fabs(e) > tol1) {
r=(x-w)*(fx-fv);
q=(x-v)*(fx-fw);
p=(x-v)*q-(x-w)*r;
q=2.0*(q-r);
if (q > 0.0)
p = -p;
q=fabs(q);
etemp=e;
e=d;
if (fabs(p) >= fabs(0.5*q*etemp) || p <= q*(a-x) || p >= q*(b-x))
d=CGOLD*(e=(x >= xm ? a-x : b-x));
else {
d=p/q;
u=x+d;
if (u-a < tol2 || b-u < tol2)
d=SIGN(tol1,xm-x);
}
} else {
d=CGOLD*(e=(x >= xm ? a-x : b-x));
}
u=(fabs(d) >= tol1 ? x+d : x+SIGN(tol1,d));
fu=computeFunction(u);
#if TRY_AITKEN
//In a Nutshell, Aitken's sequence accelerator constructs
//an auxiliary sequence, using
//"the main sequence, minus the square of the delta divided
// by the delta of the delta".
double aitken = 0;
bool haveAitken = false;
if (iter>=2) {
lastDelta = delta;
delta = u - lastU;
//For now. Later, use a calculated tolerance
if (iter>=3 && 0.00001 < fabs(delta-lastDelta) && 0.00001 < fabs(delta) ) {
haveAitken = true;
aitken = u - ( delta * delta ) / (delta - lastDelta);
}
}
cout.precision(6);
if (!haveAitken) {
cout << "f(" << u << ")=" << fu << endl;
} else {
cout << "f(" << u << ")=" << fu << ", A=" << aitken << ", f(A)=" << computeFunction(aitken) << endl;
}
lastU = u;
#endif
if (fu <= fx) {
if (u >= x)
a=x;
else
b=x;
SHFT(v,w,x,u)
SHFT(fv,fw,fx,fu)
} else {
if (u < x)
a=u;
else
b=u;
if (fu <= fw || w == x) {
v=w;
w=u;
fv=fw;
fw=fu;
} else
if (fu <= fv || v == x || v == w) {
v=u;
fv=fu;
}
}
}
*foptx = fx;
xw = x-w;
wv = w-v;
vx = v-x;
*f2optx = 2.0*(fv*xw + fx*wv + fw*vx)/(v*v*xw + x*x*wv + w*w*vx);
return x;
}
#undef ITMAX
#undef CGOLD
#undef ZEPS
#undef SHFT
#undef SIGN
#undef GOLD
#undef GLIMIT
#undef TINY
double Optimization::minimizeOneDimen(double xmin, double xguess, double xmax, double tolerance, double *fx, double *f2x) {
double eps, optx, ax, bx, cx, fa, fb, fc;
//int converged; /* not converged error flag */
/* first attempt to bracketize minimum */
if (xguess < xmin) xguess = xmin;
if (xguess > xmax) xguess = xmax;
eps = xguess*tolerance*50.0;
ax = xguess - eps;
if (ax < xmin) ax = xmin;
bx = xguess;
cx = xguess + eps;
if (cx > xmax) cx = xmax;
/* check if this works */
// compute fb first to save some computation, if any
fb = computeFunction(bx);
fa = (ax<bx) ? computeFunction(ax) : fb;
fc = (bx<cx) ? computeFunction(cx) : fb;
/* if it works use these borders else be conservative */
if ((fa < fb) || (fc < fb)) {
if (ax != xmin) fa = computeFunction(xmin);
if (cx != xmax) fc = computeFunction(xmax);
ax = xmin;
cx = xmax;
}
/*
const int MAX_ROUND = 10;
for (i = 0; ((fa < fb-tolerance) || (fc < fb-tolerance)) && (i<MAX_ROUND); i++) {
// brent method require that fb is smaller than both fa and fc
// find some random values until fb achieve this
bx = (((double)rand()) / RAND_MAX)*(cx-ax) + ax;
fb = computeFunction(bx);
}*/
/*
if ((fa < fb) || (fc < fb)) {
if (fa < fc) { bx = ax; fb = fa; } else { bx = cx; fb = fc; }
//cout << "WARNING: Initial value for Brent method is set at bound " << bx << endl;
}*/
// optx = brent_opt(xmin, xguess, xmax, tolerance, fx, f2x, fa, fb, fc);
//} else
optx = brent_opt(ax, bx, cx, tolerance, fx, f2x, fa, fb, fc);
if (*fx > fb) // if worse, return initial value
{
*fx = computeFunction(bx);
return bx;
}
return optx; /* return optimal x */
}
double Optimization::minimizeOneDimenSafeMode(double xmin, double xguess, double xmax, double tolerance, double *f)
{
double ferror;
double optx = minimizeOneDimen(xmin, xguess, xmax, tolerance, f, &ferror);
double fnew;
// check value at the boundary
if ((optx < xmax) && (fnew = computeFunction(xmax)) <= *f+tolerance) {
//if (verbose_mode >= VB_MAX)
//cout << "Note from Newton safe mode: " << optx << " (" << f << ") -> " << xmax << " ("<< fnew << ")" << endl;
optx = xmax;
*f = fnew;
}
if ((optx > xmin) && (fnew = computeFunction(xmin)) <= *f+tolerance) {
//if (verbose_mode >= VB_MAX)
//cout << "Note from Newton safe mode: " << optx << " -> " << xmin << endl;
optx = xmin;
*f = fnew;
}
return optx;
}
/*****************************************************
One dimensional optimization with Newton Raphson
only applicable if 1st and 2nd derivatives are easy to compute
*****************************************************/
double Optimization::minimizeNewtonSafeMode(double xmin, double xguess, double xmax, double tolerance, double &f)
{
double optx = minimizeNewton(xmin, xguess, xmax, tolerance, f);
double fnew;
// check value at the boundary
if ((optx < xmax) && (fnew = computeFunction(xmax)) <= f+tolerance) {
//if (verbose_mode >= VB_MAX)
//cout << "Note from Newton safe mode: " << optx << " (" << f << ") -> " << xmax << " ("<< fnew << ")" << endl;
optx = xmax;
f = fnew;
}
if ((optx > xmin) && (fnew = computeFunction(xmin)) <= f+tolerance) {
//if (verbose_mode >= VB_MAX)
//cout << "Note from Newton safe mode: " << optx << " -> " << xmin << endl;
optx = xmin;
f = fnew;
}
return optx;
}
double Optimization::minimizeNewton(double x1, double xguess, double x2, double xacc, double &d2l, int maxNRStep)
{
int j;
double df,dx,dxold,f;
double temp,xh,xl,rts, rts_old, xinit;
rts = xguess;
if (rts < x1) rts = x1;
if (rts > x2) rts = x2;
xinit = xguess;
// finit = fold = fm = computeFuncDerv(rts,f,df);
computeFuncDerv(rts,f,df);
d2l = df;
if (!isfinite(f) || !isfinite(df)) {
nrerror("Wrong computeFuncDerv");
}
if (df >= 0.0 && fabs(f) < xacc) return rts;
if (f < 0.0) {
xl = rts;
xh = x2;
} else {
xh = rts;
xl = x1;
}
dx=dxold=fabs(xh-xl);
for (j=1;j<=maxNRStep;j++) {
rts_old = rts;
if (
(df <= 0.0) // function is concave
// || (fm > fold + xacc) // increasing
|| (((rts-xh)*df-f)*((rts-xl)*df-f) >= 0.0) // out of bound
//|| (fabs(2.0*f) > fabs(dxold*df)) // converge too slow
) {
dxold=dx;
dx=0.5*(xh-xl);
rts=xl+dx;
d2l = df;
if (xl == rts) return rts;
} else {
dxold=dx;
dx=f/df;
temp=rts;
rts -= dx;
d2l = df;
if (temp == rts) return rts;
}
if (fabs(dx) < xacc || (j == maxNRStep)) {
// if (fm > finit) {
// // happen in rare cases that it is worse than starting point: revert init value
// fm = computeFunction(xinit);
// return xinit;
// }
return rts_old;
// return rts;
}
// fold = fm;
// fm = computeFuncDerv(rts,f,df);
computeFuncDerv(rts,f,df);
if (!isfinite(f) || !isfinite(df)) nrerror("Wrong computeFuncDerv");
if (df > 0.0 && fabs(f) < xacc) {
d2l = df;
// if (fm > finit) {
// // happen in rare cases that it is worse than starting point: revert init value
// fm = computeFunction(xinit);
// return xinit;
// }
return rts;
}
if (f < 0.0)
xl=rts;
else if (f > 0.0)
xh=rts;
}
nrerror("Maximum number of iterations exceeded in minimizeNewton");
d2l = 0.0;
return 0.0;
}
double Optimization::minimizeNewton(double x1, double xguess, double x2, double xacc, int maxNRStep)
{
double var;
double optx = minimizeNewton(x1, xguess, x2, xacc, var, maxNRStep);
return optx;
}
/*****************************************************
Multi dimensional optimization with Newton Raphson
only applicable if 1st and 2nd derivatives are easy to compute
*****************************************************/
int matinv (double x[], int n, int m, double space[]);
double Optimization::minimizeNewtonMulti(double *x1, double *xguess, double *x2, double xacc, int N, int maxNRStep)
{
int i, step;
int N_2 = N*N;
double df[N_2], dx[N], dxold[N], f[N+1], space[N];
double temp, xh[N], xl[N], rts[N], rts_old[N];
for (i = 0; i < N; i++) {
rts[i] = xguess[i];
if (rts[i] < x1[i]) rts[i] = x1[i];
if (rts[i] > x2[i]) rts[i] = x2[i];
}
computeFuncDervMulti(rts, f, df);
double score = f[N];
// double score = targetFunk(rts-1);
// cout << "Newton step 0: " << score << endl;
bool stop = true;
for (i = 0; i < N; i++) {
if (!(fabs(f[i]) < xacc))
stop = false;
xl[i] = x1[i];
xh[i] = x2[i];
// if (f[i] < 0.0) {
// xl[i] = rts[i];
// xh[i] = x2[i];
// } else {
// xh[i] = rts[i];
// xl[i] = x1[i];
// }
dx[i] = dxold[i] = fabs(xh[i]-xl[i]);
}
if (stop) {
memcpy(xguess, rts, sizeof(double)*N);
return score;
}
for (step = 1; step <= maxNRStep; step++) {
memcpy(rts_old, rts, sizeof(double)*N);
// first compute inverse of hessian matrix
matinv(df, N, N, space);
// now take the product
for (i = 0; i < N; i++) {
space[i] = 0.0;
for (int k = 0; k < N; k++) {
space[i] += df[i*N+k] * f[k];
}
}
// identify alpha to be not out of bound
double alpha;
for (alpha = 1.0; alpha > xacc; alpha *= 0.5) {
bool ok_bound = true;
for (i = 0; i < N; i++)
if (rts[i]-alpha*space[i] < xl[i] || rts[i]-alpha*space[i] > xh[i]) {
ok_bound = false;
break;
}
if (ok_bound) break;
}
do {
stop = true;
for (i = 0; i < N; i++) {
dxold[i] = dx[i];
dx[i] = alpha*space[i];
temp=rts[i];
rts[i] -= dx[i];
if (fabs(dx[i]) >= xacc && (step != maxNRStep)) {
stop = false;
} else {
rts[i] = rts_old[i];
}
}
double new_score = targetFunk(rts-1);
if (new_score <= score) break;
// score increased, reduce step size by half
memcpy(rts, rts_old, sizeof(double)*N);
alpha *= 0.5;
} while (!stop);
if (stop) {
break;
}
computeFuncDervMulti(rts, f, df);
score = f[N];
// double score = targetFunk(rts-1);
// cout << "Newton step " << step << ": " << score << endl;
// for (i = 0; i < N; i++) cout << " " << rts[i];
// cout << endl;
stop = true;
for (i = 0; i < N; i++) {
if (!(fabs(f[i]) < xacc)) {
stop = false;
}
// if (f[i] < 0.0)
// xl[i] = rts[i];
// else
// xh[i] = rts[i];
}
if (stop)
break;
}
// copy returned x-value
memcpy(xguess, rts, sizeof(double)*N);
if (stop)
return score;
cout << "Maximum number of iterations exceeded in minimizeNewtonMulti" << endl;
return score;
}
/*****************************************************
Multi dimensional optimization with BFGS method
*****************************************************/
#define ALF 1.0e-4
#define TOLX 1.0e-7
//static double maxarg1,maxarg2;
//#define FMAX(a,b) (maxarg1=(a),maxarg2=(b),(maxarg1) > (maxarg2) ?\
// (maxarg1) : (maxarg2))
void Optimization::lnsrch(int n, double xold[], double fold, double g[], double p[], double x[],
double *f, double stpmax, int *check, double lower[], double upper[]) {
int i;
double a,alam,alam2=0,alamin,b,disc,f2=0,fold2=0,rhs1,rhs2,slope,sum,temp,
test,tmplam;
*check=0;
for (sum=0.0,i=1;i<=n;i++) sum += p[i]*p[i];
sum=sqrt(sum);
if (sum > stpmax)
for (i=1;i<=n;i++) p[i] *= stpmax/sum;
for (slope=0.0,i=1;i<=n;i++)
slope += g[i]*p[i];
test=0.0;
for (i=1;i<=n;i++) {
temp=fabs(p[i])/max(fabs(xold[i]),1.0);
if (temp > test) test=temp;
}
alamin=TOLX/test;
alam=1.0;
bool first_time = true;
for (;;) {
// for (; alam >= alamin;) {
// bool out_of_bound = false;
// for (i=1;i<=n;i++) {
// x[i]=xold[i]+alam*p[i];
// if (x[i] < lower[i] || x[i] > upper[i]) {
// out_of_bound = true;
// break;
// }
// }
// if (out_of_bound)
// alam *= 0.5;
// else
// break;
// }
for (i=1;i<=n;i++) x[i]=xold[i]+alam*p[i];
fixBound(x, lower, upper, n);
//checkRange(x);
*f=targetFunk(x);
if (alam < alamin) {
for (i=1;i<=n;i++) x[i]=xold[i];
*check=1;
return;
} else if (*f <= fold+ALF*alam*slope) return;
else {
if (first_time)
tmplam = -slope/(2.0*(*f-fold-slope));
else {
rhs1 = *f-fold-alam*slope;
rhs2=f2-fold2-alam2*slope;
a=(rhs1/(alam*alam)-rhs2/(alam2*alam2))/(alam-alam2);
b=(-alam2*rhs1/(alam*alam)+alam*rhs2/(alam2*alam2))/(alam-alam2);
if (a == 0.0) tmplam = -slope/(2.0*b);
else {
disc=b*b-3.0*a*slope;
if (disc<0.0) //nrerror("Roundoff problem in lnsrch.");
tmplam = 0.5 * alam;
else if (b <= 0.0) tmplam=(-b+sqrt(disc))/(3.0*a);
else tmplam = -slope/(b+sqrt(disc));
}
if (tmplam>0.5*alam)
tmplam=0.5*alam;
}
}
alam2=alam;
f2 = *f;
fold2=fold;
alam=max(tmplam,0.1*alam);
first_time = false;
}
}
#undef ALF
#undef TOLX
const int MAX_ITER = 3;
extern double random_double(int *rstream);
/* changed this method to accomodate GTR20 optimization; -JD
- added new procedure called 'const' that multiplies every value by a constant value if a restart is needed
- added new option called --reset-method that determines if we do the above (const) or the old way (rand) for restarting params
- added fix to force ndim=189 on this method for GTR20 ... for some reason it likes to give 192 which makes it restart every time
- made it so if we hit the lower bound, we multiply all values up, and we multiply down if we hit the upper bound
- if both bounds are hit we go back to the old (rand) method for restarting the parameters
*/
bool Optimization::restartParameters(double guess[], int ndim, double lower[], double upper[], bool bound_check[], int iteration) {
int i;
int restart = 0;
bool aa = (ndim > 100); //lazily check if we're optimizing GTR20, which should have >100 params
if(aa) { ndim = 189; } //lazy fix -- ndim 192 -> 189 to prevent wrongful restarts
if (iteration <= MAX_ITER) {
for (i = 1; i <= ndim; i++) {
if (bound_check[i]) {
if (fabs(guess[i]-lower[i]) < 1e-4) {
// set restart to 1 (hit lower) or 3 (both) if already hit upper
if (restart == 0) { restart = 1; } else if (restart == 2) { restart = 3; }
if (restart == 3) { break; }
}
if (fabs(guess[i]-upper[i]) < 1e-4) {
// set restart to 2 (hit upper) or 3 (both) if already hit lower
if (restart == 0) { restart = 2; } else if (restart == 1) { restart = 3; }
if (restart == 3) { break; }
}
} // if bound_check
} // for
} // if iteration
if (restart > 0) {
// TODO: Dominik observed that this warning is printed even during model
// testing. A quieter solution might be preferred.
cout << "Restart estimation at the boundary... ";
// add clarification to the message printed out to console (change this if the above TODO is addressed)
bool const_mult = Params::getInstance().reset_method == "const";
if(restart == 1 && const_mult) { cout << "multiplying all values up" << endl; }
else if(restart == 2 && const_mult) { cout << "multiplying all values down" << endl; }
else { cout << "multiplying all values by random double" << endl; }
for (i = 1; i <= ndim; i++) {
// multiply by guess_multiplier if reset_method=="const" AND restart = 1 or 2 (hit upper or lower bound but not both)
// guess_multiplier is assumed to be <1
if(Params::getInstance().reset_method == "const" && restart < 3){
double val;
if(restart == 1) { // hit lower bound, multiply up
val = guess[i] * (1.0 / Params::getInstance().guess_multiplier);
if (val >= lower[i]) { guess[i] = val; }
}
if(restart == 2) { // hit upper bound, multiply down
val = guess[i] * Params::getInstance().guess_multiplier;
if (val <= upper[i]) { guess[i] = val; }
}
// use the old method if reset_method=="rand" OR restart = 3 (hit both upper and lower bounds)
} else {
guess[i] = random_double() * (upper[i] - lower[i])/3 + lower[i];
}
}
}
return (restart);
}
double Optimization::minimizeMultiDimen(double guess[], int ndim, double lower[], double upper[], bool bound_check[], double gtol, double *hessian) {
int i, iter;
double fret, minf = 1e+12;
double *minx = new double [ndim+1];
int count = 0;
bool restart;
do {
dfpmin(guess, ndim, lower, upper, gtol, &iter, &fret, hessian);
if (fret < minf) {
minf = fret;
for (i = 1; i <= ndim; i++) {
minx[i] = guess[i];
}
}
count++;
// restart the search if at the boundary
// it's likely to end at a local optimum at the boundary
restart = restartParameters(guess,ndim,lower,upper,bound_check,count);
} while (restart);
if (count > 1) {
for (i = 1; i <= ndim; i++) {
guess[i] = minx[i];
}
fret = minf;
}
delete [] minx;
return fret;
}
#define ITMAX 200
#define SQR(a) ((a)*(a))
#define EPS 3.0e-8
#define TOLX (4*EPS)
#define STPMX 100.0
#define FREEALL free_vector(xi,1,n);free_vector(pnew,1,n); \
free_matrix(hessin,1,n,1,n);free_vector(hdg,1,n);free_vector(g,1,n); \
free_vector(dg,1,n);
void Optimization::dfpmin(double p[], int n, double lower[], double upper[]
, double gtol, int *iter, double *fret, double *hessian) {
int check,i,its,j;
double den,fac,fad,fae,fp,stpmax,sum=0.0,sumdg,sumxi,temp,test;
double *dg,*g,*hdg,**hessin,*pnew,*xi;
dg=new_vector(1,n);
g=new_vector(1,n);
hdg=new_vector(1,n);
hessin=new_matrix(1,n,1,n);
pnew=new_vector(1,n);
xi=new_vector(1,n);
fp = derivativeFunk(p,g);
for (i=1;i<=n;++i)
{
xi[i] = -g[i];
sum += p[i]*p[i];
}
if (hessian) {
int pos = 0;
for (i=1; i<=n; i++) {
for (j=1; j<=n; j++, ++pos) {
hessin[i][j] = hessian[pos];
}
}
} else {
for (i=1;i<=n;i++) {
for (j=1;j<=n;j++) {
hessin[i][j]=0.0;
}
hessin[i][i]=1.0;
}
}
//checkBound(p, xi, lower, upper, n);
//checkDirection(p, xi);
stpmax=STPMX*max(sqrt(sum),(double)n);
for (its=1;its<=ITMAX;its++) {
*iter=its;
lnsrch(n,p,fp,g,xi,pnew,fret,stpmax,&check, lower, upper);
fp = *fret;
for (i=1;i<=n;i++) {
xi[i]=pnew[i]-p[i];
p[i]=pnew[i];
}
test=0.0;
for (i=1;i<=n;i++) {
temp=fabs(xi[i])/max(fabs(p[i]),1.0);
if (temp > test) test=temp;
}
if (test < TOLX) {
FREEALL
return;
}
for (i=1;i<=n;i++) dg[i]=g[i];
derivativeFunk(p,g);
test=0.0;
den=max(fabs(*fret),1.0); // fix bug found by Tung, as also suggested by NR author
for (i=1;i<=n;i++) {
temp=fabs(g[i])*max(fabs(p[i]),1.0)/den;
if (temp > test) test=temp;
}
if (test < gtol) {
FREEALL
return;
}
for (i=1;i<=n;i++) {
dg[i]=g[i]-dg[i];
}
for (i=1;i<=n;i++) {
hdg[i]=0.0;
for (j=1;j<=n;j++) {
hdg[i] += hessin[i][j]*dg[j];
}
}
fac=fae=sumdg=sumxi=0.0;
for (i=1;i<=n;i++) {
fac += dg[i]*xi[i];
fae += dg[i]*hdg[i];
sumdg += SQR(dg[i]);
sumxi += SQR(xi[i]);
}
if (fac*fac > EPS*sumdg*sumxi)
{
fac=1.0/fac;
fad=1.0/fae;
for (i=1;i<=n;i++) {
dg[i]=fac*xi[i]-fad*hdg[i];
}
for (i=1;i<=n;i++) {
for (j=1;j<=n;j++) {
hessin[i][j] += fac*xi[i]*xi[j]
-fad*hdg[i]*hdg[j]+fae*dg[i]*dg[j];
}
}
}
for (i=1;i<=n;i++) {
xi[i]=0.0;
for (j=1;j<=n;j++) xi[i] -= hessin[i][j]*g[j];
}
//checkBound(p, xi, lower, upper, n);
//checkDirection(p, xi);
//if (*iter > 200) cout << "iteration=" << *iter << endl;
}
// BQM: disable this message!
//nrerror("too many iterations in dfpmin");
FREEALL
}
#undef ITMAX
#undef SQR
#undef EPS
#undef TOLX
#undef STPMX
#undef FREEALL
#undef FMAX
/**
the approximated derivative function
@param x the input vector x
@param dfx the derivative at x
@return the function value at x
*/
double Optimization::derivativeFunk(double x[], double dfx[]) {
/*
if (!checkRange(x))
return INFINITIVE;
*/
int ndim = getNDim();
double *h = new double[ndim+1];
double temp;
int dim;
double fx = targetFunk(x);
for (dim = 1; dim <= ndim; dim++ ){
temp = x[dim];
h[dim] = ERROR_X * fabs(temp);
if (h[dim] == 0.0) h[dim] = ERROR_X;
x[dim] = temp + h[dim];
h[dim] = x[dim] - temp;
dfx[dim] = (targetFunk(x));
x[dim] = temp;
}
for (dim = 1; dim <= ndim; dim++ )
dfx[dim] = (dfx[dim] - fx) / h[dim];
delete [] h;
return fx;
}
/*#define NRANSI
#define ITMAX 100
#define CGOLD 0.3819660
#define ZEPS 1.0e-10
#define SHFT(a,b,c,d) (a)=(b);(b)=(c);(c)=(d);
#define SIGN(a,b) ((b) >= 0.0 ? fabs(a) : -fabs(a))
double Optimization::brent(double ax, double bx, double cx, double tol,
double *xmin)
{
int iter;
double a,b,d=0.0,etemp,fu,fv,fw,fx,p,q,r,tol1,tol2,u,v,w,x,xm;
double e=0.0;
a=(ax < cx ? ax : cx);
b=(ax > cx ? ax : cx);
x=w=v=bx;
fw=fv=fx=computeFunction(x);
for (iter=1;iter<=ITMAX;iter++) {
xm=0.5*(a+b);
tol2=2.0*(tol1=tol*fabs(x)+ZEPS);
if (fabs(x-xm) <= (tol2-0.5*(b-a))) {