///
/// \brief minimize starting from the initial guess x0
///
tstate operator()(const tproblem& problem, const tvector& x0) const
{
assert(problem.size() == static_cast<tsize>(x0.size()));
// current state
tstate cstate(problem, x0);
// running-weighted-averaged-per-dimension-squared gradient
average_vector_t<tscalar, tvector> gavg(x0.size());
for (tsize e = 0, k = 1; e < base_t::m_epochs; e ++)
{
for (tsize i = 0; i < base_t::m_epoch_size; i ++, k ++)
{
// learning rate
const tscalar alpha = base_t::m_alpha0;
// descent direction
gavg.update(cstate.g.array().square(), base_t::weight(k));
cstate.d = -cstate.g.array() /
(base_t::m_epsilon + gavg.value().array()).sqrt();
// update solution
cstate.update(problem, alpha);
}
base_t::ulog(cstate);
}
return cstate;
}
QVariant ActivityItem::data(int role) const
{
switch(role) {
case CodeRole:
return code();
case NameRole:
return name();
case IconRole:
return icon();
case CStateRole:
return cstate();
case BackgroundRole:
return backgrounds();
case OrderRole:
return order();
default:
return QVariant();
}
}
