virtual stats local_run_l(int index, bool learn) {
DAction* ac = fac;
double total_reward = 0;
DAction* b = new DAction(prob->getActions(), (int) false );
do
{
prob->apply(*ac);
total_reward += prob->reward();
if(learn) {
if(!prob->restrictedAction().restricted)
ac = this->computeNextAction(getState(prob), prob->reward(), prob->goal());
else {
agent->had_choosed(getState(prob), *b, prob->reward(), false, prob->goal());
ac = b;
}
// ac = this->computeNextAction(getState(prob), prob->reward(), prob->goal());
}
else {
if(ac != fac)
delete ac;
ac = this->agent->decision(getState(prob), false);
}
}
while(!prob->done());
if(!learn)
delete ac;
delete b;
return {prob->step, total_reward, prob->step, index};
}
stats local_run(int index, bool random_init) {
prob->init(random_init);
agent->startEpisode(getState(prob), *fac);
// if(no_learn_knowledge) {
local_run_l(index, true);
prob->init(random_init);
agent->startEpisode(getState(prob), *fac);
return local_run_l(index, false);
// }
// return local_run_l(index, true);
}
stats run_best(int index) {
prob->init(false);
DAction* ac = new DAction(*fac);
double total_reward = 0;
do
{
prob->apply(*ac);
delete ac;
total_reward += prob->reward();
ac = best_policy->decision(getState(prob), false);
}
while(!prob->done());
delete ac;
return {prob->step, total_reward, prob->step, index};
}
void init() {
agent = this->createAgent(getState(prob), *fac);
agentSet = true;
}