/* G02DA_A1W_F C++ Header Example Program.
*
* Copyright 2019 Numerical Algorithms Group.
* Mark 27, 2019.
*/
#include <nag.h>
#include <nagad.h>
#include <stdio.h>
#include <string>
#include <iostream>
using namespace std;
int main(void)
{
int exit_status = 0;
void *ad_handle = 0;
nagad_a1w_w_rtype tol;
Integer ifail = 0;
cout << "G02DA_A1W_F C++ Header Example Program Results\n\n";
tol.value = 0.000001;
tol.id = 0;
// Skip heading in data file
string mystr;
getline (cin, mystr);
// Read problem sizes
Integer n, m;
cin >> n;
cin >> m;
// Allocate arrays depending on m and n.
nagad_a1w_w_rtype *x=0, *y=0;
Integer *isx=0;
double *dy=0;
if (!(x = NAG_ALLOC(m*n, nagad_a1w_w_rtype)) ||
!(y = NAG_ALLOC(n, nagad_a1w_w_rtype)) ||
!(dy = NAG_ALLOC(n, double)) ||
!(isx = NAG_ALLOC(m, Integer))) {
cout << "Allocation failure\n";
exit_status = -1;
exit(exit_status);
}
// Create AD tape
nagad_a1w_ir_create();
// Create AD configuration data object
ifail = 0;
x10aa_a1w_f_(ad_handle,ifail);
// Read model data
double dd;
for (int i = 0; i<n; ++i) {
for (int j = 0; j<m; ++j) {
cin >> dd;
int k = i + j*n;
x[k].value = dd;
x[k].id = 0;
}
cin >> dd;
y[i].value = dd;
y[i].id = 0;
nagad_a1w_ir_register_variable(&y[i]);
}
// Calculate ip
Integer ip = 0;
for (int j = 0; j<m; ++j) {
cin >> isx[j];
if (isx[j]>0) {
ip++;
}
}
// Mean = 'M'
ip++;
// Allocate arrays depending on ip
nagad_a1w_w_rtype *b=0, *cov=0, *h=0, *p=0, *q=0, *res=0, *se=0, *wk=0;
double *dbdy;
Integer lcov = (ip*ip+ip)/2, lp = ip*(ip+2), lq = n*(ip+1);
Integer lwk = ip*ip + 5*(ip-1);
if (!(b = NAG_ALLOC(ip, nagad_a1w_w_rtype)) ||
!(cov = NAG_ALLOC(lcov, nagad_a1w_w_rtype)) ||
!(h = NAG_ALLOC(n, nagad_a1w_w_rtype)) ||
!(p = NAG_ALLOC(lp, nagad_a1w_w_rtype)) ||
!(q = NAG_ALLOC(lq, nagad_a1w_w_rtype)) ||
!(res = NAG_ALLOC(n, nagad_a1w_w_rtype)) ||
!(se = NAG_ALLOC(ip, nagad_a1w_w_rtype)) ||
!(wk = NAG_ALLOC(lwk, nagad_a1w_w_rtype)) ||
!(dbdy = NAG_ALLOC(n*ip, double))) {
cout << "Allocation failure\n";
exit_status = -2;
exit(exit_status);
}
// Perform Regression
Integer idf, irank;
nagad_a1w_w_rtype rss, wt[1];
logical svd;
ifail = 0;
g02da_a1w_f_(ad_handle,"M","U",n,x,n,m,isx,ip,y,wt,rss,idf,
b,se,cov,res,h,q,n,svd,irank,p,tol,wk,ifail,1,1);
// Display results
if (svd) {
cout << "Model is not of full rank, rank = " << irank << endl;
}
cout << "Residual sum of squares = " << nagad_a1w_get_value(rss);
cout << "\nDegrees of freedom = " << idf << endl;
cout << "\nVariable Parameter estimate Standard error\n\n";
cout.setf(ios::scientific,ios::floatfield);
cout.precision(2);
for (int i=0; i < ip; ++i) {
cout.width(5); cout << i << " ";
cout.width(12); cout << nagad_a1w_get_value(b[i]) << " ";
cout.width(12); cout << nagad_a1w_get_value(se[i]) << endl;
}
cout << "\n\n Derivatives calculated: First order adjoints\n";
cout << " Computational mode : algorithmic\n";
cout << "\n Derivatives:\n\n";
// Obtain derivatives
double inc = 1.0;
nagad_a1w_inc_derivative(&rss,inc);
ifail = 0;
nagad_a1w_ir_interpret_adjoint(ifail);
for (int j=0; j<n; j++) {
dy[j] = nagad_a1w_get_derivative(y[j]);
}
cout << " i d(rss)/dy(i)\n";
cout.precision(4);
for (int i=0; i<n; ++i) {
cout.width(5); cout << i << " ";
cout.width(12); cout << dy[i] << endl;
}
for (int i=0; i<ip; i++) {
// Reset adjoints, initialize derivative, and evaluate adjoint
nagad_a1w_ir_zero_adjoints();
nagad_a1w_inc_derivative(&b[i],inc);
ifail = 0;
nagad_a1w_ir_interpret_adjoint_sparse(ifail);
for (int j=0; j<n; j++) {
int k = j + i*n;
double dd = nagad_a1w_get_derivative(y[j]);
dbdy[k] = dd;
}
}
// Print matrix routine
cout << endl;
NagError fail;
INIT_FAIL(fail);
x04cac(Nag_ColMajor,Nag_GeneralMatrix,Nag_NonUnitDiag,n,ip,dbdy,n,
" Derivatives db/dy",0,&fail);
// Remove computational data object and tape
ifail = 0;
x10ab_a1w_f_(ad_handle,ifail);
nagad_a1w_ir_remove();
return exit_status;
}