double HumanAgent::evaluatePolicy(Environment<bool>& env){
#ifdef __USE_SDL
Action action;
int reward = 0;
int totalReward = 0;
int cumulativeReward = 0;
//Repeat (for each episode):
for(int episode = 0; episode < numEpisodesToEval; episode++){
int step = 0;
while(!env.game_over() && step < maxStepsInEpisode) {
action = receiveAction();
//If one wants to save trajectories, this is where the trajectory is saved:
if(toSaveTrajectory){
saveTrajectory(action);
}
reward = env.act(action);
cumulativeReward += reward;
step++;
}
printf("Episode %d, Cumulative Reward: %d\n", episode + 1, cumulativeReward);
totalReward += cumulativeReward;
cumulativeReward = 0;
env.reset_game(); //Start the game again when the episode is over
}
return double(totalReward)/numEpisodesToEval;
}
double RandomAgent::evaluatePolicy(Environment<bool>& env){
int reward = 0;
int totalReward = 0;
int cumulativeReward = 0;
int numActions;
ActionVect actions;
//Check if one wants to sample from all possible actions or only the valid ones:
if(useMinActions){
actions = env.getMinimalActionSet();
}
else{
actions = env.getLegalActionSet();
}
numActions = actions.size();
printf("Number of Actions: %d\n\n", numActions);
//Repeat (for each episode):
for(int episode = 0; episode < numEpisodesToEval; episode++){
int step = 0;
while(!env.game_over() && step < maxStepsInEpisode) {
reward = env.act(actions[rand()%numActions]);
cumulativeReward += reward;
step++;
}
printf("Episode %d, Cumulative Reward: %d\n", episode + 1, cumulativeReward);
totalReward += cumulativeReward;
cumulativeReward = 0;
env.reset_game(); //Start the game again when the episode is over
}
return double(totalReward)/numEpisodesToEval;
}