void uniformSampling(M& Result,
randEngine& mtRandom)
{
randFloat sample( mtRandom, realUniformDist(0,1) );
size_t nA = Result.size1();
size_t nB = Result.size2();
// TODO: Improve with iterators
// std::generate(Result.begin2(),Result.end2(),sample);
for (size_t i = 0; i < nA; i++)
for (size_t j = 0; j < nB; j++)
Result(i,j) = sample();
}
void lhs(M& Result, randEngine& mtRandom)
{
randFloat sample( mtRandom, realUniformDist(0,1) );
size_t nA = Result.size1();
size_t nB = Result.size2();
double ndA = static_cast<double>(nA);
// std::vector<int> perms(nA);
for (size_t i = 0; i < nB; i++)
{
// TODO: perms starts at 1. Check this
std::vector<int> perms = return_index_vector(nA);
randomPerms(perms, mtRandom);
for (size_t j = 0; j < nA; j++)
{
Result(j,i) = ( static_cast<double>(perms[j]) - sample() ) / ndA;
}
}
}
int GP_Hedge::update_hedge()
{
// We just care about the differences
double max_l = *std::max_element(loss_.begin(),loss_.end());
loss_ += svectord(loss_.size(),max_l);
// To avoid overflow
double mean_g = std::accumulate(gain_.begin(),gain_.end(),0.0)
/ static_cast<double>(gain_.size());
gain_ -= svectord(gain_.size(),mean_g);
// Optimal eta according to Shapire
double max_g = *std::max_element(gain_.begin(),gain_.end());
double eta = (std::min)(10.0,sqrt(2.0*log(3.0)/max_g));
// Compute probabilities
std::transform(gain_.begin(), gain_.end(), prob_.begin(),
boost::bind(softmax,_1,eta));
//Normalize
double sum_p =std::accumulate(prob_.begin(),prob_.end(),0.0);
prob_ /= sum_p;
//Update bandits gain
gain_ -= loss_;
std::partial_sum(prob_.begin(), prob_.end(), cumprob_.begin(),
std::plus<double>());
randFloat sampleUniform( *mtRandom, realUniformDist(0,1));
double u = sampleUniform();
for (size_t i=0; i < cumprob_.size(); ++i)
{
if (u < cumprob_(i))
return i;
}
FILE_LOG(logERROR) << "Error updating Hedge algorithm. "
<< "Selecting first criteria by default.";
return 0;
};
GP_Hedge::GP_Hedge():
mtRandom(100u),
sampleUniform( mtRandom, realUniformDist(0,1))
{};
