/
memoization.h
1614 lines (1484 loc) · 54 KB
/
memoization.h
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
#ifndef __MEMOIZATION_H__
#define __MEMOIZATION_H__
#pragma syntax nocheck
#add_scan(memoFun)
#include <vector>
#include <map>
#include <set>
#include <math.h>
#include <stdlib.h>
using namespace std;
int __max_history_size = 100;
void set_memoization_max_history(int v) {
__max_history_size = v;
}
template<class T>
bool is_less(T * A, T * B, int n) {
for (int i = 0; i < n; i++)
if (A[i] < B[i])
return true;
else if (A[i] == B[i])
continue;
else
return false;
return false;
}
template<class T>
bool is_equal(T * A, T * B, int n) {
for (int i = 0; i < n; i++)
if (A[i] == B[i])
continue;
else
return false;
return true;
}
template<class T>
void fill_into(vector<double> * V, T * A, int n) {
for (int i = 0; i < n; i++)
V->push_back(A[i]);
}
template<class T>
void fill_out(vector<double> * V, int & ptr, T * A, int n) {
for (int i = 0; i < n; i++)
A[i] = (*V)[ptr++];
}
template<class T>
double diff(T * A, T * B, int n) {
double result = 0.0;
for (int i = 0; i < n; i++)
result += fabs(A[i] - B[i]);
return result;
}
class __Predictor {
protected:
vector< vector<double> * > XX;
vector< vector<double> * > YY;
public:
__Predictor() {
XX.reserve(1000);
YY.reserve(1000);
}
virtual ~__Predictor() {
clear();
}
virtual void clear() {
for (auto & X : XX) {
delete X;
}
XX.clear();
XX.reserve(1000);
for (auto & Y : YY) {
delete Y;
}
YY.clear();
YY.reserve(1000);
}
virtual void addPointX(vector<double> * X) {
if (XX.size() == XX.capacity()) XX.reserve(XX.size() + 200);
XX.push_back(X);
}
virtual void addPointY(vector<double> * Y) {
if (YY.size() == YY.capacity()) YY.reserve(YY.size() + 200);
YY.push_back(Y);
}
virtual void calculate_sample(vector<double> X, vector<double> & Y) = 0;
virtual void train() = 0;
};
class __FeedForwardNet : public __Predictor {
private:
double tol;
double nu;
int MAX_EPOCHS;
vector< double > MMAX, MMIN, YMAX, YMIN;
vector<char> funs;
vector< vector<double> > B;
vector< vector< vector<double> > > K;
vector< vector<double> > X1, Y1;
inline double S(char mode, double s) {
if (mode == 'e')
return 1.0 / (1.0 + exp(-s));
else if (mode == 'h')
return tanh(s);
else if (mode == 'l')
return s;
else if (mode == 'r')
return s < 0.0 ? 0.0 : s;
else
throw runtime_error("Unknown activation function");
}
inline double D(char mode, double s, double f_s) {
if (mode == 'e')
return f_s*(1.0 - f_s);
else if (mode == 'h')
return 1.0 - f_s*f_s;
else if (mode == 'l')
return 1.0;
else if (mode == 'r')
return s < 0.0 ? 0.0 : 1.0;
else
throw runtime_error("Unknown activation function");
}
public:
__FeedForwardNet(double tol, double nu, int max_epochs, int nx, int ny, const vector<int> & layers, const vector<char> & funs) : __Predictor() {
this->tol = tol;
this->nu = nu;
MAX_EPOCHS = max_epochs;
this->funs = funs;
B.resize(layers.size()+1);
K.resize(layers.size()+1);
X1.resize(layers.size()+1);
Y1.resize(layers.size()+1);
for (int i = 0; i <= layers.size(); i++) {
int nw = i == 0 ? nx : layers[i-1];
int nn = i < layers.size() ? layers[i] : ny;
B[i].resize(nn);
X1[i].resize(nn);
Y1[i].resize(nn);
K[i].resize(nn);
for (int j = 0; j < nn; j++) {
B[i][j] = 1.0*rand()/RAND_MAX;
K[i][j].resize(nw);
for (int k = 0; k < nw; k++)
K[i][j][k] = 1.0*rand()/RAND_MAX;
}
}
}
void calculate_model(vector<double> * X) {
if (X1.size() != B.size()) X1.resize(B.size());
if (Y1.size() != B.size()) Y1.resize(B.size());
for (int i = 0; i < B.size(); i++) {
if (X1[i].size() != B[i].size()) X1[i].resize(B[i].size());
if (Y1[i].size() != B[i].size()) Y1[i].resize(B[i].size());
for (int j = 0; j < B[i].size(); j++) {
double y = B[i][j];
if (i == 0)
for (int k = 0; k < X->size(); k++)
y += K[i][j][k]*(*X)[k];
else
for (int k = 0; k < B[i-1].size(); k++)
y += K[i][j][k]*Y1[i-1][k];
X1[i][j] = y;
Y1[i][j] = i == B.size()-1 ? y : S(funs[i], y);
}
}
}
virtual void calculate_sample(vector<double> X, vector<double> & Y) {
for (int p = 0; p < X.size(); p++) {
double d = MMAX[p]==MMIN[p] ? 1.0 : 2.0/(MMAX[p]-MMIN[p]);
X[p] = -1.0 + (X[p]-MMIN[p])*d;
}
calculate_model(&X);
Y.resize(B[B.size()-1].size());
for (int p = 0; p < Y.size(); p++) {
double d = YMAX[p]==YMIN[p] ? 1.0 : 2.0/(YMAX[p]-YMIN[p]);
Y[p] = YMIN[p] + (Y1[B.size()-1][p]+1.0)/d;
}
}
virtual void train() {
int L = XX.size();
int M = B.size();
if (L == 0) return;
MMAX.resize(XX[0]->size());
MMIN.resize(XX[0]->size());
for (int p = 0; p < XX[0]->size(); p++) {
MMAX[p] = -1E100;
MMIN[p] = 1E100;
for (int i = 0; i < L; i++) {
if ((*XX[i])[p] > MMAX[p]) MMAX[p] = (*XX[i])[p];
if ((*XX[i])[p] < MMIN[p]) MMIN[p] = (*XX[i])[p];
}
}
YMAX.resize(YY[0]->size());
YMIN.resize(YY[0]->size());
for (int p = 0; p < YY[0]->size(); p++) {
YMAX[p] = -1E100;
YMIN[p] = 1E100;
for (int i = 0; i < L; i++) {
if ((*YY[i])[p] > YMAX[p]) YMAX[p] = (*YY[i])[p];
if ((*YY[i])[p] < YMIN[p]) YMIN[p] = (*YY[i])[p];
}
}
for (int p = 0; p < XX[0]->size(); p++) {
double d = MMAX[p]==MMIN[p] ? 1.0 : 2.0/(MMAX[p]-MMIN[p]);
for (int i = 0; i < L; i++)
(*XX[i])[p] = -1.0 + ((*XX[i])[p]-MMIN[p])*d;
}
for (int p = 0; p < YY[0]->size(); p++) {
double yd = YMAX[p]==YMIN[p] ? 1.0 : 2.0/(YMAX[p]-YMIN[p]);
for (int i = 0; i < L; i++)
(*YY[i])[p] = -1.0 + ((*YY[i])[p]-YMIN[p])*yd;
}
vector< vector<double> > DB(M);
vector< vector< vector<double> > > DK(M);
const double alpha = 0.2;
int epochs = 0;
double err;
for (int i = 0; i < M; i++) {
DB[i].resize(B[i].size());
DK[i].resize(B[i].size());
for (int j = 0; j < B[i].size(); j++)
DK[i][j].resize(i == 0 ? XX[0]->size() : B[i-1].size());
}
vector<int> nums(L);
for (int i = 0; i < L; i++)
nums[i] = i;
vector<int> idxs(L);
for (int i = 0; i < L; i++) {
int idx = (int)(1.0*(nums.size()-1)*rand()/RAND_MAX);
idxs[i] = nums[idx];
nums.erase(nums.begin()+idx);
}
do {
err = 0.0;
for (int ii = 0; ii < L; ii++) {
int i = idxs[ii];
calculate_model(XX[i]);
for (int yy = 0; yy < B[M-1].size(); yy++) {
double YS = Y1[M-1][yy];
double Y = (*YY[i])[yy];
double delta = YS-Y;
err += fabs(delta);
DB[M-1][yy] = alpha*DB[M-1][yy] + (1-alpha)*(-nu*delta);
B[M-1][yy] += DB[M-1][yy];
if (M == 1) {
for (int j = 0; j < XX[0]->size(); j++)
for (int k = 0; k < B[0].size(); k++) {
DK[0][k][j] = alpha*DK[0][k][j] + (1-alpha)*(-nu*delta*(*XX[i])[j]);
K[0][k][j] += DK[0][k][j];
}
} else {
for (int k = 0; k < DK[M-1][yy].size(); k++) {
DK[M-1][yy][k] = alpha*DK[M-1][yy][k] + (1-alpha)*(-nu*delta*Y1[M-2][k]);
K[M-1][yy][k] += DK[M-1][yy][k];
}
static double * deltas = new double[65536*4];
memset(deltas, 0, 65536*4*sizeof(double));
for (int k = 0; k < DK[M-1][yy].size(); k++) {
deltas[k] = D(funs[M-2], X1[M-2][k], Y1[M-2][k])*delta*K[M-1][yy][k];
}
for (int lr = M-2; lr > 0; lr--) {
for (int j = 0; j < B[lr-1].size(); j++)
for (int k = 0; k < B[lr].size(); k++) {
DK[lr][k][j] = alpha*DK[lr][k][j] + (1 - alpha)*(-nu*deltas[k] * Y1[lr-1][j]);
K[lr][k][j] += DK[lr][k][j];
}
for (int k = 0; k < B[lr].size(); k++) {
DB[lr][k] = alpha*DB[lr][k] + (1 - alpha)*(-nu*deltas[k]);
B[lr][k] += DB[lr][k];
}
static double * deltas1 = new double[65536*4];
memset(deltas1, 0, 65536*4*sizeof(double));
for (int j = 0; j < B[lr-1].size(); j++)
for (int k = 0; k < B[lr].size(); k++) {
deltas1[j] += D(funs[lr-1], X1[lr-1][j], Y1[lr-1][j])*deltas[k] * K[lr][k][j];
}
for (int j = 0; j < B[lr-1].size(); j++)
deltas[j] = deltas1[j];
}
for (int j = 0; j < XX[0]->size(); j++)
for (int k = 0; k < B[0].size(); k++) {
DK[0][k][j] = alpha*DK[0][k][j] + (1-alpha)*(-nu*deltas[k]*(*XX[i])[j]);
K[0][k][j] += DK[0][k][j];
}
for (int k = 0; k < B[0].size(); k++) {
DB[0][k] = alpha*DB[0][k] + (1-alpha)*(-nu*deltas[k]);
B[0][k] += DB[0][k];
}
}
}
}
// err /= L;
err /= B[M-1].size();
} while (err > 0.05*tol && epochs++ < MAX_EPOCHS);
}
};
#ifndef max
#define max(A,B) ((A) > (B) ? (A) : (B))
#endif
#ifndef min
#define min(A,B) ((A) < (B) ? (A) : (B))
#endif
#if _POSIX_C_SOURCE >= 200112L
#define _isnan isnan
#endif
namespace solve {
/* LU - разложение с выбором максимального элемента по диагонали */
bool _GetLU(int NN, int * iRow, long double * A, long double * LU)
{
int i,j,k;
try {
memmove(LU,A,NN*NN*sizeof(long double));
for (i=0;i<NN;i++)
iRow[i]=i;
for (i=0;i<NN-1;i++)
{
long double Big = fabs(LU[iRow[i]*NN+i]);
int iBig = i;
long double Kf;
for (j=i+1;j<NN;j++)
{
long double size = fabs(LU[iRow[j]*NN+i]);
if (size>Big)
{
Big = size;
iBig = j;
}
}
if (iBig!=i)
{
int V = iRow[i];
iRow[i] = iRow[iBig];
iRow[iBig] = V;
}
Kf = 1.0/LU[iRow[i]*NN+i];
LU[iRow[i]*NN+i] = Kf;
for (j=i+1;j<NN;j++)
{
long double Fact = Kf*LU[iRow[j]*NN+i];
LU[iRow[j]*NN+i] = Fact;
for (k=i+1;k<NN;k++)
LU[iRow[j]*NN+k] -= Fact*LU[iRow[i]*NN+k];
}
}
LU[(iRow[NN-1]+1)*NN-1] = 1.0/LU[(iRow[NN-1]+1)*NN-1];
}
catch (...) {
return false;
}
return true;
}
/* Метод LU - разложения */
bool _SolveLU(int NN, int * iRow, long double * LU, long double * Y, long double * X)
{
int i,j,k;
try {
X[0] = Y[iRow[0]];
for (i=1;i<NN;i++)
{
long double V = Y[iRow[i]];
for (j=0;j<i;j++)
V -= LU[iRow[i]*NN+j]*X[j];
X[i] = V;
}
X[NN-1]*=LU[(iRow[NN-1]+1)*NN-1];
for (i=1,j=NN-2;i<NN;i++,j--)
{
long double V = X[j];
for (k=j+1;k<NN;k++)
V -= LU[iRow[j]*NN+k]*X[k];
X[j] = V*LU[iRow[j]*NN+j];
}
}
catch (...) {
return false;
}
return true;
}
}
// Расчет параметров функции Y = f(X1,X2) = g+a*X1+b*X2+c*X1*X2+d*X1^2+e*X2^2. Метод наименьших квадратов
// Вычисляет также YY -- значения выходов, вычисленные по f(X1,X2). Возвращает вектор (g, a, b, c, d, e)
long double * MNK_of_X1X2(const std::vector<double> X1, const std::vector<double> X2,
const std::vector<double> Y, double &err,
std::vector<double> & YY) {
const int N = 6;
int x1mask[N] = {0,1,0,1,2,0};
int x2mask[N] = {0,0,1,1,0,2};
int Z = X1.size();
std::vector<std::vector<long double>> XP1(5,std::vector<long double>(Z,1.0));
std::vector<std::vector<long double>> XP2(5,std::vector<long double>(Z,1.0));
long double * A = new long double[N*N];
long double * LU = new long double[N*N];
long double * B = new long double[N];
long double * XX = new long double[N];
int * iLU = new int[N];
int i,j,k, ptr;
long double Q;
YY.resize(Z,0.0);
for (i=1; i<5; i++)
for (j=0; j<Z; j++) {
XP1[i][j] = XP1[i-1][j]*X1[j];
XP2[i][j] = XP2[i-1][j]*X2[j];
}
for (i=0, ptr=0; i<N; i++) {
for (j=0; j<N; j++, ptr++) {
Q = 0.0;
for (k=0; k<Z; k++)
Q += XP1[x1mask[j]][k]*XP2[x2mask[j]][k]*XP1[x1mask[i]][k]*XP2[x2mask[i]][k];
A[ptr] = Q;
}
Q = 0.0;
for (k=0; k<Z; k++)
Q += Y[k]*XP1[x1mask[i]][k]*XP2[x2mask[i]][k];
B[i] = Q;
}
if (!(solve::_GetLU(N, iLU, A, LU) && solve::_SolveLU(N, iLU, LU, B, XX))) {
memset(XX,0,N*sizeof(long double));
err = 1E300;
} else {
for (k=0; k<N; k++)
if (_isnan(XX[k])) {
memset(XX,0,N*sizeof(long double));
err = 1E300;
return XX;
}
err = 0.0;
for (j=0; j<Z; j++) {
long double R = 0.0;
for (k=N-1; k>=0; k--)
R += XP1[x1mask[k]][j]*XP2[x2mask[k]][j]*XX[k];
YY[j] = R;
double cur_err = (R-Y[j])*(R-Y[j]);
err += cur_err;
}
}
delete[] A;
delete[] LU;
delete[] B;
delete[] iLU;
return XX;
}
// Вычисление коэффициента корреляции векторов X и Y. Попутно считает среднеквадратичные отклонения SX, SY
double correlation(const std::vector<double> X, const std::vector<double> Y, double & SX, double & SY) {
double corr = 0.0;
double avrX = 0.0;
double avrY = 0.0;
int N = X.size();
for (int i=0; i<N; i++) {
avrX += X[i];
avrY += Y[i];
}
avrX /= N;
avrY /= N;
SX = SY = 0.0;
for (int i=0; i<N; i++) {
double vx = X[i]-avrX;
double vy = Y[i]-avrY;
SX += vx*vx;
SY += vy*vy;
corr += vx*vy;
}
SX = sqrt(SX/(N-1));
SY = sqrt(SY/(N-1));
return corr/(N-1);
}
class MGUA_X1X2;
class MGUA_X1X2 {
public:
long double MGUAkoeffs[6];
int refcount;
bool FirstLayer;
union {
struct {
MGUA_X1X2 * X1;
MGUA_X1X2 * X2;
};
struct {
double * NX1; double Min1, Max1;
double * NX2; double Min2, Max2;
};
};
MGUA_X1X2(long double * Kf, MGUA_X1X2 * _X1, MGUA_X1X2 * _X2):
X1(_X1), X2(_X2) {
FirstLayer = false; refcount = 0;
memmove(MGUAkoeffs,Kf,sizeof(MGUAkoeffs));
X1->refcount++;
X2->refcount++;
delete[] Kf;
};
MGUA_X1X2(long double * Kf, double * _X1, double * _X2, double min1, double max1, double min2, double max2):
NX1(_X1), NX2(_X2), Min1(min1), Max1(max1), Min2(min2), Max2(max2) {
FirstLayer = true; refcount = 0;
memmove(MGUAkoeffs,Kf,sizeof(MGUAkoeffs));
delete[] Kf;
};
virtual double getValue(const bool restrict = false) {
double _X1, _X2;
if (FirstLayer) {
double _NX1 = restrict ? (min(max(Min1, *NX1),Max1)) : *NX1;
double _NX2 = restrict ? (min(max(Min2, *NX2),Max2)) : *NX2;
_X1 = fabs(Max1-Min1)>1E-5 ? (_NX1-Min1)/(Max1-Min1) : _NX1;
_X2 = fabs(Max2-Min2)>1E-5 ? (_NX2-Min2)/(Max2-Min2) : _NX2;
} else {
_X1 = X1->getValue(restrict);
_X2 = X2->getValue(restrict);
}
return MGUAkoeffs[0]+MGUAkoeffs[1]*_X1+MGUAkoeffs[2]*_X2+
MGUAkoeffs[3]*_X1*_X2+MGUAkoeffs[4]*_X1*_X1+MGUAkoeffs[5]*_X2*_X2;
}
~MGUA_X1X2() {
if (!FirstLayer) {
if ((--(X1->refcount))==0) delete X1;
if ((--(X2->refcount))==0) delete X2;
}
}
};
class TModelMGUA {
private:
MGUA_X1X2 * STRUCT;
double denorm_value;
public:
TModelMGUA(double EPS, double * X, const std::vector<std::vector<double>> &XX, const std::vector<double> &YY, double &err);
TModelMGUA(MGUA_X1X2 * out) {
STRUCT = out;
STRUCT->refcount++;
};
virtual double getValue(const bool restrict = false) {
return STRUCT ? STRUCT->getValue(restrict) : denorm_value;
};
virtual ~TModelMGUA() {
if (STRUCT && (--(STRUCT->refcount))==0)
delete STRUCT;
}
};
typedef struct {
std::vector<double> YY;
MGUA_X1X2 * ref;
} variantMGUA;
// Поиск интерполятора с помощью МГУА. XX[Z][N] -- объекты, N -- кол-во, Z -- число координат.
// YY[N] -- отклики. _X[Z] -- входы.
TModelMGUA::TModelMGUA(double EPS,
double * _X,
const std::vector< std::vector<double> > & XX,
const std::vector<double> & YY,
double &err) {
STRUCT = NULL;
int N = YY.size();
int Z = XX.size();
std::vector<double> mins(Z), maxs(Z);
std::map<double,int> correlations;
double SY;
for (int i=0; i<Z; i++) {
double SX;
double corr = correlation(XX[i],YY,SX,SY);
if (fabs(SY)<EPS)
correlations[0.0] = -1;
else if (fabs(SX*SY)>EPS) {
double CORR = fabs(corr/(SX*SY));
std::map<double,int>::iterator itc = correlations.find(CORR);
if (itc==correlations.end()) correlations[CORR] = i;
else if (++itc==correlations.end()) correlations[CORR+0.000001] = i;
else correlations[(CORR+itc->first)*0.5] = i;
}
double xMin = 1E300;
double xMax = -1E300;
for (int j=0; j<N; j++) {
if (XX[i][j]<xMin) xMin = XX[i][j];
if (XX[i][j]>xMax) xMax = XX[i][j];
}
mins[i] = xMin;
maxs[i] = xMax;
}
err = 1E300;
if (correlations.size()>0) {
std::map<double,int>::iterator itc0, itc = correlations.end();
int BestV = (--itc)->second;
if (BestV>=0) {
std::map<double,variantMGUA> MGUA;
// ATTENTION! In Metalayer there is another formala
int NPolies = max(1,(int)(0.8*(correlations.size()-1)*correlations.size()));
// END
double LastCriterion = 1E300;
int i;
for (itc = correlations.begin(), i = 0; itc!=correlations.end() && i<NPolies; itc++, i++) {
int V1 = itc->second;
std::vector<double> VV(N);
variantMGUA var;
for (int j=0; j<N; j++)
if (maxs[V1]-mins[V1]>EPS)
VV[j] = (XX[V1][j]-mins[V1])/(maxs[V1]-mins[V1]);
else
VV[j] = XX[V1][j];
var.YY = VV;
var.ref = reinterpret_cast<MGUA_X1X2 *>(V1);
MGUA[100.0-itc->first] = var; // Чтобы переменные с большей корреляцией с выходами попали в начало популяции.
}
while (MGUA.size()>1) {
std::map<double,variantMGUA>::iterator i0,i1;
std::map<double,variantMGUA> _MGUA;
variantMGUA var;
for (i0 = MGUA.begin(); i0!=MGUA.end(); i0++)
for (i1 = i0, i1++; i1!=MGUA.end(); i1++) {
if (reinterpret_cast<long long>(i0->second.ref) < Z) // В MGUA -- фиктивный первый слой (входные переменные)
var.ref = new MGUA_X1X2(MNK_of_X1X2(i0->second.YY,i1->second.YY,YY,err,var.YY),
_X+reinterpret_cast<long long>(i0->second.ref),_X+reinterpret_cast<long long>(i1->second.ref),
mins[reinterpret_cast<long long>(i0->second.ref)],maxs[reinterpret_cast<long long>(i0->second.ref)],
mins[reinterpret_cast<long long>(i1->second.ref)],maxs[reinterpret_cast<long long>(i1->second.ref)]);
else // В MGUA -- настоящий слой
var.ref = new MGUA_X1X2(MNK_of_X1X2(i0->second.YY,i1->second.YY,YY,err,var.YY),i0->second.ref,i1->second.ref);
if (err<1E300) _MGUA[err] = var;
else delete var.ref;
}
MGUA.clear();
i0 = _MGUA.begin();
if (i0!=_MGUA.end()) {
int newNPolies = max(1,(double)min(NPolies,(int)_MGUA.size())*0.9);
err = i0->first;
if (LastCriterion-err<=1E-2 || newNPolies<2) {
MGUA[err] = i0->second;
if (!STRUCT) {
STRUCT = i0->second.ref;
STRUCT->refcount++;
}
i0++;
while (i0!=_MGUA.end())
delete (i0++)->second.ref;
break;
} else {
i1 = _MGUA.begin();
for (int k=0; k<NPolies && i1!=_MGUA.end(); k++) i1++;
MGUA.insert(_MGUA.begin(),i1);
while (i1!=_MGUA.end())
delete (i1++)->second.ref;
}
NPolies = newNPolies;
LastCriterion = err;
}
}
}
}
if (err == 1E300) {
denorm_value = 0.0;
for (double d : YY)
denorm_value += d;
if (YY.size() > 0) denorm_value /= YY.size();
err = 0.0;
for (double d : YY)
err += (d-denorm_value)*(d-denorm_value);
}
}
class __MGUA : public __Predictor {
private:
vector<TModelMGUA *> predictors;
int NX, NY;
double * _X;
double EPS;
public:
__MGUA(double tol, int nx, int ny) : __Predictor() {
_X = new double[nx];
NX = nx;
NY = ny;
EPS = tol;
}
virtual void clear() {
__Predictor::clear();
for (TModelMGUA * &p : predictors)
delete p;
predictors.clear();
}
~__MGUA() {
delete[] _X;
}
virtual void calculate_sample(vector<double> X, vector<double> & Y) {
for (int x = 0; x < NX; x++)
_X[x] = X[x];
Y.resize(NY);
for (int y = 0; y < NY; y++)
Y[y] = predictors[y]->getValue();
}
virtual void train() {
int N = XX.size();
for (TModelMGUA * &p : predictors)
delete p;
predictors.clear();
int Z = NX;
int K = NY;
// _XX[Z][N] -- объекты, N -- кол-во, Z -- число координат.
// _YY[N] -- отклики. _X[Z] -- входы.
vector< vector<double> > _XX(Z);
vector<double> _YY(N);
for (int z = 0; z < Z; z++) {
_XX[z].resize(N);
for (int n = 0; n < N; n++)
_XX[z][n] = XX[n]->at(z);
}
predictors.resize(K);
double err;
for (int k = 0; k < K; k++) {
for (int n = 0; n < N; n++)
_YY[n] = YY[n]->at(k);
predictors[k] = new TModelMGUA(EPS, _X, _XX, _YY, err);
}
}
};
class __LIN_EXTRAPOLATOR : public __Predictor {
private:
int NY;
int P;
double EPS;
int * WP;
double ** K;
long double * A;
long double * LU;
long double * Y;
long double * X;
public:
__LIN_EXTRAPOLATOR(double tol, int p, int ny) : __Predictor() {
P = p;
NY = ny;
WP = new int[NY];
K = new double * [NY];
for (int i = 0; i < NY; i++)
K[i] = new double[P];
A = new long double[P*P];
LU = new long double[P*P];
Y = new long double[P];
X = new long double[P];
EPS = tol;
}
~__LIN_EXTRAPOLATOR() {
delete[] WP;
for (int i = 0; i < NY; i++)
delete[] K[i];
delete[] K;
delete[] A;
delete[] LU;
delete[] Y;
delete[] X;
}
virtual void addPointX(vector<double> * X) {
if (XX.size() > 5*P) {
delete *XX.begin();
XX.erase(XX.begin());
}
XX.push_back(X);
}
virtual void addPointY(vector<double> * Y) {
if (YY.size() > 5*P) {
delete *YY.begin();
YY.erase(YY.begin());
}
YY.push_back(Y);
}
virtual void calculate_sample(vector<double> X, vector<double> & Y) {
map<double, int> timeline;
int * IDX = new int[XX.size()];
for (int i = 0; i < XX.size(); i++)
timeline[XX[i]->at(0)] = i;
map<double, int>::iterator it = timeline.begin();
for (int c = 0; it != timeline.end(); it++, c++) {
IDX[c] = it->second;
}
Y.resize(NY);
for (int y = 0; y < NY; y++) {
double val = K[y][0];
for (int j = 1; j < WP[y]; j++)
val += K[y][j] * YY[IDX[XX.size() - WP[y] + j]]->at(y);
Y[y] = val;
}
delete[] IDX;
}
virtual void train() {
map<double, int> timeline;
long double * VV = new long double[XX.size()];
for (int i = 0; i < XX.size(); i++)
timeline[XX[i]->at(0)] = i;
int iRow[1024];
for (int p = 0; p < NY; p++) {
map<double, int>::iterator it = timeline.begin();
for (int c = 0; it != timeline.end(); it++, c++) {
VV[c] = YY[it->second]->at(p);
}
double max_err = 1E300;
WP[p] = -1;
for (int pp = P; pp >= 1; pp--) {
memset(A, 0, P*P*sizeof(long double));
memset(Y, 0, P*sizeof(long double));
int L = XX.size();
for (int i = max(0, L - 1.5*P); i < L - pp + 1; i++) {
for (int j = 0, ptr = 0; j < pp; j++) {
long double XXV = j == 0 ? 1.0 : (long double) VV[i + j - 1];
for (int k = 0; k < pp; k++, ptr++) {
A[ptr] += (k == 0 ? 1.0 : (long double) VV[i + k - 1])*XXV;
}
Y[j] += XXV*(long double) VV[i + pp - 1];
}
}
bool result = _SolveGaussZeidel(pp, iRow, A, LU, Y, X) || _GetLU(pp, iRow, A, LU) && _SolveLU(pp, iRow, LU, Y, X);
if (result) {
for (int i = 0, ptr = 0; result && i < pp; i++) {
double val = 0.0;
for (int j = 0; result && j < pp; j++, ptr++) {
if (X[j] != X[j]) result = false;
val += A[ptr] * X[j];
}
double r = fabs(val - Y[i]);
if (r != r || r > 1E-3)
result = false;
}
}
double cur_err = 1E300;
if (result) {
int n = 0;
cur_err = 0.0;
for (int i = max(0, L - 1.5*P); i < L - pp + 1; i++, n++) {
double val = X[0];
for (int j = 1; j < pp; j++)
val += X[j] * ((double) VV[i + j - 1]);
double H = VV[i + pp - 1];
if (H != 0) {
cur_err += fabs(1.0*(H - val) / H);
} else if (val != 0)
cur_err += fabs(1.0*(H - val) / val);
}
if (n > 0) cur_err /= n;
}
if (result) {
if (cur_err <= max_err) {
WP[p] = pp;
for (int i = 0; i < pp; i++)
K[p][i] = (double)X[i];
max_err = cur_err;
}
}
}
}
delete[] VV;
}
};
#def_pattern memoFun => make_memoization(gid(), //FIRST_MEMO/@Value, //OTHER_MEMOS/@Value, //MODE/@Value, //FIRST_DESC/@Value, //OTHER_DESCS/@Value, //FIRST_REF/@Value, //FIRST_TYPE/@Value, //FIRST/@Value, //FIRST_IDXS/@Value, //OTHER_REF/@Value, //OTHER_TYPE/@Value, //OTHER/@Value, //OTHER_IDXS/@Value, //THREEP/@Value, /root/ARGS/@Value, /root/RETTYPE/@Value, /root/ID/@Value, //CONTROLLED/@Value, //C1/@Value, //C2/@Value) {
(((^)|(\;)+|\}|\\n)((\s|\\t)*\\n)*)(\s|\\t)*
@begin
\#pragma(\s|\\t)+memoization(\s|\\t)*
\((\s|\\t)*((\*)*[giot])->{FIRST_MEMO}((\s|\\t)*\,(\s|\\t)*((\*)*[giot])->{OTHER_MEMOS})*\)
(
(\s|\\t)*
([A-Za-z0-9_]+)->{MODE}
(\s|\\t)*
\((\s|\\t)*([A-Za-z0-9_\*\.\+\-]+)->{FIRST_DESC}((\s|\\t)*\,(\s|\\t)*([A-Za-z0-9_\*\.\+\-]+)->{OTHER_DESCS})*\)
(
(
(\s|\\t)*
(controlled)->{CONTROLLED}
(\s|\\t)*
\((\s|\\t)*([A-Za-z0-9_]+)->{C1}(\s|\\t)*\,(\s|\\t)*([A-Za-z0-9_]+)->{C2}(\s|\\t)*\)
)|
(
(\s|\\t)*
(conditions)->{CONTROLLED}
(\s|\\t)*
\((\s|\\t)*((.{1,255})->{C1}\,)?=>{Predicates.BAL($,',')}
(\s|\\t)*((.{1,255})->{C2}\))?=>{Predicates.BAL($,')')}(\s|\\t)*
)
)?
)?
(\s|\\t)*\\n
(\s|\\t)*
((([A-Za-z0-9_\[\]\*]+((\\n|\s|\\t)*\*(\\n|\s|\\t)*|(\\n|\s|\\t)+))+)->{RETTYPE}(\w+(\\n|\s|\\t)*)->{ID}\(
(
((((\\n|\s|\\t)*[A-Za-z0-9_\[\]]+)+)->{FIRST_TYPE}
(
((\\n|\s|\\t)*(\&|(\*)+)->{FIRST_REF}
)|
(\\n|\s|\\t)+()->{FIRST_REF}
)