Type objective_function<Type>::operator() ()
{
// Data
DATA_VECTOR( y_i );
DATA_MATRIX( X_ij );
// Parameters
PARAMETER_VECTOR( b_j );
PARAMETER_VECTOR( theta_z );
// Objective funcction
Type zero_prob = 1 / (1 + exp(-theta_z(0)));
Type logsd = exp(theta_z(1));
Type jnll = 0;
int n_data = y_i.size();
// Linear predictor
vector<Type> linpred_i( n_data );
linpred_i = X_ij * b_j;
// Probability of data conditional on fixed effect values
for( int i=0; i<n_data; i++){
if(y_i(i)==0) jnll -= log( zero_prob );
if(y_i(i)!=0) jnll -= log( 1-zero_prob ) + dlognorm( y_i(i), linpred_i(i), logsd, true );
}
// Reporting
REPORT( zero_prob );
REPORT( logsd );
REPORT( linpred_i );
return jnll;
}
std::chrono::duration<double> cphf(calculation_data &data){
std::chrono::duration<double> elapsed_seconds(0);
auto start = std::chrono::steady_clock::now();
int sz = (data.n_paired/2)*(data.n_baseset-data.n_paired/2);
std::function<int(int,int)> idx = [&](int i,int a){
return (data.n_baseset-data.n_paired/2)*i + (a-data.n_paired/2);
};
// build A + B matrix for singlet
hermitian_matrix AB = data.A_singlet + data.B_singlet;
hermitian_matrix inv_AB = AB.inverse();
// build theta matrix
matrix theta_x(sz,1);
matrix theta_y(sz,1);
matrix theta_z(sz,1);
for(int i=0;i<data.n_paired/2;i++){
for(int a=data.n_paired/2;a<data.n_baseset;a++){
int ia = idx(i,a);
theta_x(ia,0) = data.mo_dipole_x(i,a);
theta_y(ia,0) = data.mo_dipole_y(i,a);
theta_z(ia,0) = data.mo_dipole_z(i,a);
}
}
// solve (A+B)C=Theta
matrix Cx = inv_AB * theta_x;
matrix Cy = inv_AB * theta_y;
matrix Cz = inv_AB * theta_z;
// calculate the polarizability
matrix *Cs[3] = { &Cx,&Cy,&Cz };
matrix *thetas[3] = { &theta_x,&theta_y,&theta_z };
for(int xyz1=0;xyz1<3;xyz1++)
for(int xyz2=0;xyz2<3;xyz2++)
data.polarizability[xyz1][xyz2] = 0;
for(int i=0;i<data.n_paired/2;i++){
for(int a=data.n_paired/2;a<data.n_baseset;a++){
int ia = idx(i,a);
for(int xyz1=0;xyz1<3;xyz1++)
for(int xyz2=0;xyz2<3;xyz2++)
data.polarizability[xyz1][xyz2] += 4 * (*thetas[xyz1])(ia,0) * (*Cs[xyz2])(ia,0);
}
}
auto end = std::chrono::steady_clock::now();
elapsed_seconds = std::chrono::duration_cast<std::chrono::duration<double>>(end-start);
return elapsed_seconds;
}
Type objective_function<Type>::operator() ()
{
// Data
DATA_INTEGER( like ); // define likelihood type, 1==delta lognormal, 2==delta gamma
DATA_VECTOR( y_i ); // observations
DATA_MATRIX( X_ij ); // covariate design matrix
DATA_VECTOR( include ); //0== include in NLL, 1== exclude from NLL
// Parameters
PARAMETER_VECTOR( b_j ); // betas to generate expected values
PARAMETER_VECTOR( theta_z ); // variances
// Transformations
Type zero_prob = 1 / (1 + exp(-theta_z(0)));
Type sd = exp(theta_z(1)); //standard deviation (lognormal), scale parameter theta (gamma)
int n_data = y_i.size();
Type jnll = 0;
Type pred_jnll = 0;
vector<Type> jnll_i(n_data);
// linear predictor
vector<Type> logpred_i( n_data );
logpred_i = X_ij * b_j;
// Delta lognormal
if(like==1){
for( int i=0; i<n_data; i++){
if(y_i(i)==0) jnll_i(i) -= log( zero_prob );
if(y_i(i)!=0) jnll_i(i) -= log( 1-zero_prob ) + dlognorm( y_i(i), logpred_i(i), sd, true );
// Running counter
if( include(i)==0 ) jnll += jnll_i(i);
if( include(i)==1 ) pred_jnll += jnll_i(i);
}
}
// Delta gamma
if(like==2){
for(int i=0; i<n_data; i++){
if(y_i(i)==0) jnll_i(i) -= log( zero_prob );
if(y_i(i)!=0) jnll_i(i) -= log( 1-zero_prob ) + dgamma( y_i(i), pow(sd,-2), exp(logpred_i(i))*pow(sd,2), true );
// Running counter
if( include(i)==0 ) jnll += jnll_i(i);
if( include(i)==1 ) pred_jnll += jnll_i(i);
}
}
// Reporting
REPORT( zero_prob );
REPORT( sd );
REPORT( logpred_i );
REPORT( b_j );
REPORT( pred_jnll );
REPORT( jnll_i );
return jnll;
}
