double MarkovModel::loglike()const{
const Vec &icount(suf()->init());
const Mat &tcount(suf()->trans());
Vec logpi0(log(pi0()));
Mat logQ(log(Q()));
double ans= icount.dot(logpi0);
ans+= el_mult_sum(tcount, logQ);
return ans;
}
double MarkovModel::loglike(const Vector &serialized_params)const{
const Vec &icount(suf()->init());
const Mat &tcount(suf()->trans());
int S = state_space_size();
TransitionProbabilityMatrix transition_probabilities(S);
Vec logpi0(log(pi0()));
Mat logQ(log(Q()));
double ans= icount.dot(logpi0);
ans+= el_mult_sum(tcount, logQ);
return ans;
}
Type objective_function<Type>::operator() ()
{
DATA_INTEGER(minAge);
DATA_INTEGER(maxAge);
DATA_INTEGER(minYear);
DATA_INTEGER(maxYear);
DATA_ARRAY(catchNo);
DATA_ARRAY(stockMeanWeight);
DATA_ARRAY(propMature);
DATA_ARRAY(M);
DATA_INTEGER(minAgeS);
DATA_INTEGER(maxAgeS);
DATA_INTEGER(minYearS);
DATA_INTEGER(maxYearS);
DATA_SCALAR(surveyTime);
DATA_ARRAY(Q1);
PARAMETER_VECTOR(logN1Y);
PARAMETER_VECTOR(logN1A);
PARAMETER_VECTOR(logFY);
PARAMETER_VECTOR(logFA);
PARAMETER_VECTOR(logVarLogCatch);
PARAMETER_VECTOR(logQ);
PARAMETER(logVarLogSurvey);
int na=maxAge-minAge+1;
int ny=maxYear-minYear+1;
int nas=maxAgeS-minAgeS+1;
int nys=maxYearS-minYearS+1;
// setup F
matrix<Type> F(ny,na);
for(int y=0; y<ny; ++y){
for(int a=0; a<na; ++a){
F(y,a)=exp(logFY(y))*exp(logFA(a));
}
}
// setup logN
matrix<Type> logN(ny,na);
for(int a=0; a<na; ++a){
logN(0,a)=logN1Y(a);
}
for(int y=1; y<ny; ++y){
logN(y,0)=logN1A(y-1);
for(int a=1; a<na; ++a){
logN(y,a)=logN(y-1,a-1)-F(y-1,a-1)-M(y-1,a-1);
if(a==(na-1)){
logN(y,a)=log(exp(logN(y,a))+exp(logN(y,a-1)-F(y-1,a)-M(y-1,a)));
}
}
}
matrix<Type> predLogC(ny,na);
for(int y=0; y<ny; ++y){
for(int a=0; a<na; ++a){
predLogC(y,a)=log(F(y,a))-log(F(y,a)+M(y,a))+log(Type(1.0)-exp(-F(y,a)-M(y,a)))+logN(y,a);
}
}
Type ans=0;
for(int y=0; y<ny; ++y){
for(int a=0; a<na; ++a){
if(a==0){
ans+= -dnorm(log(catchNo(y,a)),predLogC(y,a),exp(Type(0.5)*logVarLogCatch(0)),true);
}else{
ans+= -dnorm(log(catchNo(y,a)),predLogC(y,a),exp(Type(0.5)*logVarLogCatch(1)),true);
}
}
}
matrix<Type> predLogS(nys,nas);
for(int y=0; y<nys; ++y){
for(int a=0; a<nas; ++a){
int sa = a+(minAgeS-minAge);
int sy = y+(minYearS-minYear);
predLogS(y,a) = logQ(a)-(F(sy,sa)+M(sy,sa))*surveyTime+logN(sy,sa);
ans += -dnorm(log(Q1(y,a)),predLogS(y,a),exp(Type(0.5)*logVarLogSurvey),true);
}
}
vector<Type> ssb(ny);
ssb.setZero();
for(int y=0; y<=ny; ++y){
for(int a=0; a<na; ++a){
std::cout<<y<<" "<<a<<" "<<"\n";
ssb(y)+=exp(logN(y,a))*stockMeanWeight(y,a)*propMature(y,a);
}
}
ADREPORT(ssb);
return ans;
}
