int main()
{
int nvar=20; // This is the number of independent variables
independent_variables x(1,nvar); // these are the independent variables
double f; // This is the dependent variable
dvector g(1,nvar); // Holds the vector of partial derivatives (the gradient)
fmm fmc(nvar); // Create structure to manage minimization
BEGIN_MINIMIZATION(nvar,f,x,g,fmc) // Macro to set up beginning of
// minimization loop
f=fcomp(nvar,x);
END_MINIMIZATION(nvar,g) // Macro to set up end of minimization loop
cout << " The minimizing values are\n" << x << "\n"; //Print out the answer
return 0;
}
void print_count(int width, int height)
{
mpz_t count;
mpz_init(count);
fmc(&count, width, height);
size_t optimal_bits = mpz_sizeinbase(count, 2);
int naive_bits = (width-1)*height + width*(height-1); /* i.e. using 1 bit per edge of the graph */
gmp_printf("There are %Zd different mazes on a %dx%d grid. "
"That’s a %zd-bit number, compared with %d bits for a naive packing, a saving of %.2f%%.\n",
count, width, height, optimal_bits, naive_bits, 100.0 * (1.0 - (float) optimal_bits / naive_bits));
mpz_clear(count);
}
void GLControlWidget::drawText()
{
glPushAttrib(GL_LIGHTING_BIT | GL_TEXTURE_BIT);
glDisable(GL_LIGHTING);
glDisable(GL_TEXTURE_2D);
qglColor(white);
QString str("Rendering text in OpenGL is easy with Qt");
QFontMetrics fm(font());
renderText((width() - fm.width(str)) / 2, 15, str);
QFont f("courier", 8);
QFontMetrics fmc(f);
qglColor(QColor("skyblue"));
int x, y, z;
x = (xRot >= 0) ? (int) xRot % 360 : 359 - (QABS((int) xRot) % 360);
y = (yRot >= 0) ? (int) yRot % 360 : 359 - (QABS((int) yRot) % 360);
z = (zRot >= 0) ? (int) zRot % 360 : 359 - (QABS((int) zRot) % 360);
str.sprintf("Rot X: %03d - Rot Y: %03d - Rot Z: %03d", x, y, z);
renderText((width() - fmc.width(str)) / 2, height() - 15, str, f);
glPopAttrib();
}
/**
* Description not yet available.
* \param
*/
void function_minimizer::trust_region_update(int nvar,int _crit,
independent_variables& x,const dvector& _g,const double& _f)
{
double & f= (double&)_f;
dvector & g= (dvector&)_g;
fmm fmc(nvar);
if (random_effects_flag)
{
initial_params::set_active_only_random_effects();
//int unvar=initial_params::nvarcalc(); // get the number of active
initial_params::restore_start_phase();
initial_params::set_inactive_random_effects();
int nvar1=initial_params::nvarcalc(); // get the number of active
if (nvar1 != nvar)
{
cerr << "failed sanity check in "
"void function_minimizer::quasi_newton_block" << endl;
ad_exit(1);
}
}
uistream uis("admodel.hes");
if (!uis)
{
cerr << "Error trying to open file admodel.hes" << endl;
ad_exit(1);
}
int hnvar = 0;
uis >> hnvar;
dmatrix Hess(1,hnvar,1,hnvar);
uis >> Hess;
if (!uis)
{
cerr << "Error trying to read Hessian from admodel.hes" << endl;
ad_exit(1);
}
dmatrix tester(1,hnvar,1,hnvar);
tester=Hess;
dvector e=sort(eigenvalues(Hess));
double lambda=-e(1)+100.;
for (int i=1;i<=hnvar;i++)
{
tester(i,i)+=lambda;
}
dvector step = x-solve(tester,g);
{
// calculate the number of random effects unvar
// this turns on random effects variables and turns off
// everything else
//cout << nvar << endl;
initial_params::set_active_only_random_effects();
//cout << nvar << endl;
int unvar=initial_params::nvarcalc(); // get the number of active
//df1b2_gradlist::set_no_derivatives();
if (lapprox)
{
delete lapprox;
lapprox=0;
df1b2variable::pool->deallocate();
for (int i=0;i<df1b2variable::adpool_counter;i++)
{
delete df1b2variable::adpool_vector[i];
df1b2variable::adpool_vector[i]=0;
df1b2variable::nvar_vector[i]=0;
df1b2variable::adpool_counter=0;
}
}
lapprox=new laplace_approximation_calculator(nvar,unvar,1,nvar+unvar,
this);
initial_df1b2params::current_phase=initial_params::current_phase;
initial_df1b2params::save_varsptr();
allocate();
initial_df1b2params::restore_varsptr();
df1b2_gradlist::set_no_derivatives();
int ynvar=initial_params::nvarcalc_all();
dvector y(1,ynvar);
initial_params::xinit_all(y);
initial_df1b2params::reset_all(y);
g=(*lapprox)(step,f,this);
}
} // end block for quasi newton minimization
/**
* Description not yet available.
* \param
*/
void laplace_approximation_calculator::generate_antithetical_rvs()
{
// number of random vectors
const ivector & itmp=(*num_local_re_array)(1,num_separable_calls);
//const ivector & itmpf=(*num_local_fixed_array)(1,num_separable_calls);
for (int i=2;i<=num_separable_calls;i++)
{
if (itmp(i) != itmp(i-1))
{
cerr << "At present can only use antithetical rv's when "
"all separable calls are the same size" << endl;
ad_exit(1);
}
}
int n=itmp(1);
int samplesize=num_importance_samples;
// mesh size
double delta=0.01;
// maximum of distribution is near here
double mid=sqrt(double(n-1));
dmatrix weights(1,2*n,1,2);
double spread=15;
if (mid-spread<=0.001)
spread=mid-0.1;
double ssum=0.0;
double x=0.0;
double tmax=(n-1)*log(mid)-0.5*mid*mid;
for (x=mid-spread;x<=mid+spread;x+=delta)
{
ssum+=exp((n-1)*log(x)-0.5*x*x-tmax);
}
double tsum=0;
dvector dist(1,samplesize+1);
dist.initialize();
int is=0;
int ii;
for (x=mid-spread;x<=mid+spread;x+=delta)
{
tsum+=exp((n-1)*log(x)-0.5*x*x-tmax)/ssum*samplesize;
int ns=int(tsum);
for (ii=1;ii<=ns;ii++)
{
dist(++is)=x;
}
tsum-=ns;
}
if (is==samplesize-1)
{
dist(samplesize)=mid;
}
else if (is<samplesize-1)
{
cerr << "This can't happen" << endl;
exit(1);
}
// get random numbers
random_number_generator rng(rseed);
if (antiepsilon)
{
if (allocated(*antiepsilon))
{
delete antiepsilon;
antiepsilon=0;
}
}
antiepsilon=new dmatrix(1,samplesize,1,n);
dmatrix & M=*antiepsilon;
M.fill_randn(rng);
for (int i=1;i<=samplesize;i++)
{
M(i)=M(i)/norm(M(i));
}
int nvar=(samplesize-1)*n;
independent_variables xx(1,nvar);
ii=0;
for (int i=2;i<=samplesize;i++)
{
for (int j=1;j<=n;j++)
{
xx(++ii)=M(i,j);
}
}
fmmt1 fmc(nvar,5);
//fmm fmc(nvar,5);
fmc.noprintx=1;
fmc.iprint=10;
fmc.maxfn=2500;
fmc.crit=1.e-6;
double f;
double fbest=1.e+50;;
dvector g(1,nvar);
dvector gbest(1,nvar);
dvector xbest(1,nvar);
gbest.fill_seqadd(1.e+50,0.);
{
while (fmc.ireturn>=0)
{
//int badflag=0;
fmc.fmin(f,xx,g);
if (fmc.ihang)
{
//int hang_flag=fmc.ihang;
//double maxg=max(g);
//double ccrit=fmc.crit;
//int current_ifn=fmc.ifn;
}
if (fmc.ireturn>0)
{
f=fcomp1(xx,dist,samplesize,n,g,M);
if (f < fbest)
{
fbest=f;
gbest=g;
xbest=xx;
}
}
}
xx=xbest;
}
ii=0;
for (int i=2;i<=samplesize;i++)
{
for (int j=1;j<=n;j++)
{
M(i,j)=xx(++ii);
}
}
for (int i=1;i<=samplesize;i++)
{
M(i)*=dist(i)/norm(M(i));
}
}
