Symbol get_column(SEXP arg, const Environment& env, const LazySubsets& subsets ){
Symbol res = extract_column(arg, env) ;
if( !subsets.count(res) ){
stop("result of column() expands to a symbol that is not a variable from the data: %s", CHAR(PRINTNAME(res)) ) ;
}
return res ;
}
/**
* Compute the covvariance array for AR1 case
* This is based on the parameter rho.
* This is done in two steps:
* 1) fist calculate the correlation matrix dCor given phi
* 2) Get the mle estimate of m_sig2 given the correlation.
**/
void logistic_normal::compute_covariance_matrix(const dvariable& phi)
{
m_V.allocate(m_y1,m_y2,m_b1,m_nB2-1,m_b1,m_nB2-1);
m_V.initialize();
dvar3_array dCor(m_y1,m_y2,m_b1,m_nB2,m_b1,m_nB2);
dCor.initialize();
int i,j,k,nb;
for( i = m_y1; i <= m_y2; i++ )
{
nb = m_nB2(i);
dCor(i) = identity_matrix(m_b1,nb);
// 2). Compute the vector of coefficients.
dvar_vector drho(m_b1,nb);
for( j = m_b1; j <= nb; j++ )
{
drho(j) = pow(phi,j-m_b1+1);
}
// 3). Compute correlation matrix dCor
for( j = m_b1; j <= nb; j++ )
{
for( k = m_b1; k <= nb; k++ )
{
if( j != k ) dCor(i)(j,k) = drho(m_b1+abs(j-k));
}
}
m_V(i) = trans(trans(dCor(i).sub(m_b1,nb-1)).sub(m_b1,nb-1));
}
// compute mle estimate of sigma
compute_mle_sigma(m_V);
// cout<<"Got to here sigma = "<<m_sig<<endl;
for( i = m_y1; i <= m_y2; i++ )
{
nb = m_nB2(i);
for( j = m_b1; j <= nb; j++ )
{
dCor(i).rowfill(j, extract_row(dCor(i),j)*m_sig );
}
for( k = m_b1; k <= nb; k++ )
{
dCor(i).colfill(k, extract_column(dCor(i),k)*m_sig );
}
//cout<<dCor(i)<<endl;
// Kmat
dmatrix I = identity_matrix(m_b1,nb-1);
dmatrix tKmat(m_b1,nb,m_b1,nb-1);
tKmat.sub(m_b1,nb-1) = I;
tKmat(nb) = -1;
dmatrix Kmat = trans(tKmat);
m_V(i) = Kmat * dCor(i) * tKmat;
}
}
/**
* Compute the covvariance array for AR2 case
* This is based on the parameters rho and psi.
* Note that phi1 is bounded (-1,1)
* Then set phi2 = -1 + (1 + |phi1|)*psi
* and psi is bounded (0,1)
* This implies the upper bound for phi2 is 1.0 when |phi1|= 1 and psi = 1
**/
void logistic_normal::compute_covariance_matrix(const dvariable& phi1, const dvariable& psi)
{
m_V.allocate(m_y1,m_y2,m_b1,m_nB2-1,m_b1,m_nB2-1);
m_V.initialize();
dvar3_array dCor(m_y1,m_y2,m_b1,m_nB2,m_b1,m_nB2);
dCor.initialize();
int i,j,k,nb;
for( i = m_y1; i <= m_y2; i++ )
{
nb = m_nB2(i);
// dmatrix I = identity_matrix(m_b1,nb-1);
// m_V(i) = I;
dCor(i) = identity_matrix(m_b1,nb);
// 2). Compute the vector of coefficients.
dvariable phi2 = -1. + (1. + sfabs(phi1)) * psi;
dvar_vector drho(m_b1-1,nb-1);
drho = 1.0;
drho(m_b1) = phi1 / (1.0 - phi2);
for( j = m_b1+1; j <= nb-1; j++ )
{
drho(j) = phi1*drho(j-1) + phi2*drho(j-2);
}
// 3). Compute correlation matrix m_V
for( j = m_b1; j < nb; j++ )
{
for( k = m_b1; k < nb; k++ )
{
//if( j != k ) m_V(i)(j,k) = drho(m_b1+abs(j-k));
if( j != k ) dCor(i)(j,k) = drho(m_b1+abs(j-k));
}
}
m_V(i) = trans(trans(dCor(i).sub(m_b1,nb-1)).sub(m_b1,nb-1));
}
// compute mle estimate of sigma
compute_mle_sigma(m_V);
for( i = m_y1; i <= m_y2; i++ )
{
nb = m_nB2(i);
for( j = m_b1; j < nb; j++ )
{
m_V(i).rowfill(j, extract_row(m_V(i),j) * m_sig );
}
for( k = m_b1; k < nb; k++ )
{
m_V(i).colfill(k, extract_column( m_V(i),k ) * m_sig );
}
// Kmat
dmatrix I = identity_matrix(m_b1,nb-1);
dmatrix tKmat(m_b1,nb,m_b1,nb-1);
tKmat.sub(m_b1,nb-1) = I;
tKmat(nb) = -1;
dmatrix Kmat = trans(tKmat);
m_V(i) = Kmat * dCor(i) * tKmat;
}
}
