real_T, 2U> &fx, const struct13_T *DREAMPar, const ::coder::

array<real_T, 2U> &ratInputs_problemStruct_fitLimits, const ::

coder::array<real_T, 2U> &ratInputs_priors, ::coder::array<

real_T, 1U> &log_L, ::coder::array<real_T, 1U> &log_PR)

{

::coder::array<real_T, 2U> PR;

::coder::array<real_T, 2U> b_x;

int32_T i;

int32_T loop_ub;

int32_T loop_ub_tmp;

// Now calculate the likelihood (not used) and log-likelihood (used)

// ---------------------------------------

// For RAT, all the calculations are of the mvnpdf type, so remove the

// other options.

// ------------------------------------ AVH

// % If number of measurements larger than 0 --> simulation

// if Meas_info.N > 0

//

// % Initialize "res" (residual matrix)

// res = NaN(Meas_info.N,DREAMPar.N);

//

// % Loop over each model realization

// for ii = 1 : DREAMPar.N

//

// % We now calculate the error residual

// res(:,ii) = (Meas_info.Y(:) - fx(1:Meas_info.N,ii));

//

// end

//

// else

//

// % Do nothing, fx is a density or log-density returned by the PDF handle

//

// end

// ----------------------- Calculate log-prior ----------------------------

// No ABC --> regular priors (pdfs)

// if ~DREAMPar.ABC

//

// % Calculate the log-prior

// if isfield(Par_info,'prior_marginal')

//

// % Compute prior densities for each parameter in each sequence

// for qq = 1 : DREAMPar.d

// for zz = 1 : DREAMPar.N

// % Compute prior density of proposal

// PR(zz,qq) = max ( eval(char(strrep(Par_info.prior_marginal(qq),'rnd(','pdf(x(zz,qq),'))) , 1e-299 );

// end

// end

//

// % Take the log of the densities and their sum

// log_PR = sum ( log ( PR ) , 2 );

//

// elseif isfield(Par_info,'mvnpdf')

// RAT specific prior funtion (mvnpdf)

loop_ub_tmp = static_cast<int32_T>(DREAMPar->N);

PR.set_size(1, loop_ub_tmp);

for (i = 0; i < loop_ub_tmp; i++) {

PR[i] = 0.0;

}

if (loop_ub_tmp - 1 >= 0) {

loop_ub = x.size(1);

}

// Take log of any non-zero values..

log_PR.set_size(loop_ub_tmp);

for (int32_T b_i{0}; b_i < loop_ub_tmp; b_i++) {

// Loop over all the chains..

b_x.set_size(1, x.size(1));

for (i = 0; i < loop_ub; i++) {

b_x[i] = x[b_i + x.size(0) * i];

}

PR[b_i] = scaledGaussPrior(b_x, ratInputs_problemStruct_fitLimits,

ratInputs_priors);

// mvnpdf automatically goes over all pars

log_PR[b_i] = 0.0;

}

i = PR.size(1);

for (int32_T b_i{0}; b_i < i; b_i++) {

real_T d;

d = PR[b_i];

if (d != 0.0) {

log_PR[b_i] = d;

// log(PR(i)); % Does it even need to be log?

} else {

// Otherwise keep the zero values

log_PR[b_i] = 0.0;

}

}

// log_PR = sum(log_PR(:)); % Enforce column vector

// else

// No use of prior --> set log-prior to zero (no effect in Metropolis)

// log_PR = zeros ( DREAMPar.N , 1 );

//

// end

//

//

// else

//

// Diagnostic Bayes --> if summary metric is defined as prior

// if isfield(DREAMPar,'prior_handle')

//

// Evaluate distance between observed and simulated summary metrics

// for ii = 1 : DREAMPar.N

//

// Calculate summary metrics for "fx"

// S_sim = DREAMPar.prior_handle ( fx(:,ii) );

//

// Now calculate log-density (not a true log-density! - but does not matter)

// log_PR(ii,1) = max ( abs ( Meas_info.S(:) - S_sim(:) ) );

//

// end

//

// Regular ABC with summary metrics as likelihood function

// else

//

// log_PR = zeros ( DREAMPar.N , 1 );

//

// end

//

// end

// --------------------- End Calculate log-prior ---------------------------

// -------------------- Calculate log-likelihood ---------------------------

// Loop over each model realization and calculate log-likelihood of each fx

log_L.set_size(loop_ub_tmp);

for (i = 0; i < loop_ub_tmp; i++) {

log_L[i] = 0.0;

}

for (int32_T ii{0}; ii < loop_ub_tmp; ii++) {

log_L[ii] = fx[ii];

}

// ------------------ End Calculate log-likelihood -------------------------