double FidoInterface::calcObjective(double alpha, double beta, double gamma) {
std::vector<std::vector<std::string> > names;
std::vector<double> probs, empq, estq;
double roc ,mse, objective;
proteinGraph_->setAlphaBetaGamma(alpha, beta, gamma);
proteinGraph_->getProteinProbs();
proteinGraph_->getProteinProbsAndNames(names, probs);
getEstimated_and_Empirical_FDR(names, probs, empq, estq);
getROC_AUC(names, probs, roc);
getFDR_MSE(estq, empq, mse);
objective = (kObjectiveLambda * roc) - fabs((1-kObjectiveLambda) * mse);
if (VERB > 2) {
std::cerr.precision(10);
std::cerr << "Grid searching Alpha= " << alpha <<
" Beta= " << beta <<
" Gamma= " << gamma << std::endl;
std::cerr.unsetf(std::ios::floatfield);
std::cerr << "The ROC AUC estimated values is : " << roc << std::endl;
std::cerr << "The MSE FDR estimated values is : " << mse << std::endl;
std::cerr << "Objective function with second roc and mse is : " <<
objective << std::endl;
}
return objective;
}
void FidoInterface::gridSearchOptimize()
{
if(VERB > 1)
{
std::cerr << "Running super grid search..." << std::endl;
}
double gamma_best, alpha_best, beta_best;
gamma_best = alpha_best = beta_best = -1.0;
double best_objective = -100000000;
std::vector<std::vector<std::string> > names;
std::vector<double> probs,empq,estq;
double roc,mse,current_objective;
double alpha_step = 0.05;
double beta_step = 0.05;
double gamma_step = 0.05;
double beta_init = 0.00001;
double alpha_init = 0.001;
double gamma_init = 0.1;
double gamma_limit = 0.5;
double beta_limit = 0.05;
double alpha_limit = 0.5;
if(alpha != -1)
{
alpha_init = alpha_limit = alpha;
}
if(beta != -1)
{
beta_init = beta_limit = beta;
}
if(gamma != -1)
{
gamma_init = gamma_limit = gamma;
}
//NOTE very annoying the residue error of the floats that get acummulated in every iteration
for (double i = gamma_init; i <= gamma_limit; i+=gamma_step)
{
double gamma_local = i;
for (double j = log10(beta_init); j <= Round(log10(beta_limit),2); j+=beta_step)
{
double original = pow(10,j);
double beta_local = original - beta_init;
if(beta_local > 0.0) beta_local = original;
for (double k = log10(alpha_init); k <= Round(log10(alpha_limit),2); k+=alpha_step)
{
double alpha_local = pow(10,k);
proteinGraph->setAlphaBetaGamma(alpha_local, beta_local, gamma_local);
proteinGraph->getProteinProbs();
proteinGraph->getProteinProbsAndNames(names,probs);
getEstimated_and_Empirical_FDR(names,probs,empq,estq);
getROC_AUC(names,probs,roc);
getFDR_MSE(estq,empq,mse);
current_objective = (lambda * roc) - fabs(((1-lambda) * (mse)));
if(VERB > 2)
{
std::cerr.precision(10);
std::cerr << "Grid searching Alpha= " << alpha_local << " Beta= " << beta_local << " Gamma= " << gamma_local << std::endl;
std::cerr.unsetf(std::ios::floatfield);
std::cerr << "The ROC AUC estimated values is : " << roc << std::endl;
std::cerr << "The MSE FDR estimated values is : " << mse << std::endl;
std::cerr << "Objective function with second roc and mse is : " << current_objective << std::endl;
}
if (current_objective > best_objective)
{
best_objective = current_objective;
gamma_best = gamma_local;
alpha_best = alpha_local;
beta_best = beta_local;
}
}
}
}
alpha = alpha_best;
beta = beta_best;
gamma = gamma_best;
}
void FidoInterface::gridSearch()
{
double gamma_best, alpha_best, beta_best;
gamma_best = alpha_best = beta_best = -1.0;
double best_objective = -100000000;
std::vector<std::vector<std::string> > names;
std::vector<double> probs,empq,estq;
std::vector<long double> gamma_search,beta_search,alpha_search;
double roc, mse,current_objective;
switch(depth)
{
case 0:
gamma_search = boost::assign::list_of(0.5);
beta_search = boost::assign::list_of(0.001);
alpha_search = boost::assign::list_of(0.008)(0.032)(0.128);
break;
case 1:
gamma_search = boost::assign::list_of(0.1)(0.5)(0.9);
beta_search = boost::assign::list_of(0.001);
for (double k = 0.002; k <= 0.4; k*=4)
{
alpha_search.push_back(k);
}
break;
case 2:
gamma_search = boost::assign::list_of(0.1)(0.3)(0.5)(0.75)(0.9);
beta_search = boost::assign::list_of(0.001);
for (double k = 0.001; k <= 0.4; k*=2)
{
alpha_search.push_back(k);
}
break;
default:
gamma_search = boost::assign::list_of(0.5);
beta_search = boost::assign::list_of(0.001);
alpha_search = boost::assign::list_of(0.008)(0.032)(0.128);
}
if(alpha != -1)
alpha_search = boost::assign::list_of(alpha);
if(beta != -1)
beta_search = boost::assign::list_of(beta);
if(gamma != -1)
gamma_search = boost::assign::list_of(gamma);
for (unsigned int i = 0; i < gamma_search.size(); i++)
{
double gamma_local = gamma_search[i];
for (unsigned int j = 0; j < alpha_search.size(); j++)
{
double alpha_local = alpha_search[j];
for (unsigned int k = 0; k < beta_search.size(); k++)
{
double beta_local = beta_search[k];
proteinGraph->setAlphaBetaGamma(alpha_local, beta_local, gamma_local);
proteinGraph->getProteinProbs();
proteinGraph->getProteinProbsAndNames(names,probs);
getEstimated_and_Empirical_FDR(names,probs,empq,estq);
getROC_AUC(names,probs,roc);
getFDR_MSE(estq,empq,mse);
current_objective = (lambda * roc) - fabs(((1-lambda) * (mse)));
if(VERB > 2)
{
std::cerr.precision(10);
std::cerr << "Grid searching Alpha= " << alpha_local << " Beta= " << beta_local << " Gamma= " << gamma_local << std::endl;
std::cerr.unsetf(std::ios::floatfield);
std::cerr << "The ROC AUC estimated values is : " << roc << std::endl;
std::cerr << "The MSE FDR estimated values is : " << mse << std::endl;
std::cerr << "Objective function with second roc and mse is : " << current_objective << std::endl;
}
if (current_objective > best_objective)
{
best_objective = current_objective;
gamma_best = gamma_local;
alpha_best = alpha_local;
beta_best = beta_local;
}
}
}
}
alpha = alpha_best;
beta = beta_best;
gamma = gamma_best;
}