int main(int argc, char** argv) {
if (argc < 2) {
std::cerr << "Usage: " << argv[0] << " rom_file" << std::endl;
return 1;
}
ALEInterface ale;
// Get & Set the desired settings
ale.setInt("random_seed", 123);
//The default is already 0.25, this is just an example
ale.setFloat("repeat_action_probability", 0.25);
#ifdef __USE_SDL
ale.setBool("display_screen", true);
ale.setBool("sound", true);
#endif
// Load the ROM file. (Also resets the system for new settings to
// take effect.)
ale.loadROM(argv[1]);
// Get the vector of legal actions
ActionVect legal_actions = ale.getLegalActionSet();
// Play 10 episodes
for (int episode=0; episode<1; episode++) {
float totalReward = 0;
while (!ale.game_over()) {
/*
const ALEScreen screen = ale.getScreen();
float row_sum = 0, column_sum = 0, tot = 0.0;
for (int i = 0; i < screen.height(); ++i)
{
for (int j = 0; j < screen.width(); ++j)
{
pixel_t tmp = screen.get(i, j);
if(tmp >= 41 && tmp <= 43) {
row_sum += i;
column_sum += j;
tot++;
}
}
}
cout << (row_sum/tot) << "," << (column_sum/tot) << endl;
*/
Action a = legal_actions[rand() % legal_actions.size()];
// Apply the action and get the resulting reward
float reward = ale.act(a);
totalReward += reward;
}
cout << "Episode " << episode << " ended with score: " << totalReward << endl;
ale.reset_game();
}
return 0;
}
int main(int argc, char** argv) {
if (argc < 2) {
std::cout << "Usage: " << argv[0] << " rom_file" << std::endl;
return 1;
}
ALEInterface ale;
// Get & Set the desired settings
ale.setInt("random_seed", 123);
// We enable both screen and sound, which we will need for recording.
ale.setBool("display_screen", true);
// You may leave sound disabled (by setting this flag to false) if so desired.
ale.setBool("sound", true);
std::string recordPath = "record";
std::cout << std::endl;
// Set record flags
ale.setString("record_screen_dir", recordPath.c_str());
ale.setString("record_sound_filename", (recordPath + "/sound.wav").c_str());
// We set fragsize to 64 to ensure proper sound sync
ale.setInt("fragsize", 64);
// Not completely portable, but will work in most cases
std::string cmd = "mkdir ";
cmd += recordPath;
system(cmd.c_str());
// Load the ROM file. (Also resets the system for new settings to
// take effect.)
ale.loadROM(argv[1]);
// Get the vector of legal actions
ActionVect legal_actions = ale.getLegalActionSet();
// Play a single episode, which we record.
while (!ale.game_over()) {
Action a = legal_actions[rand() % legal_actions.size()];
// Apply the action (discard the resulting reward)
ale.act(a);
}
std::cout << std::endl;
std::cout << "Recording complete. To create a video, you may want to run \n"
" doc/scripts/videoRecordingExampleJoinXXX.sh. See manual for details.." << std::endl;
return 0;
}
Learner::Learner(ALEInterface& ale, Parameters *param) : bproFeatures(param->gameName){
delta = 0.0;
cumReward = 0;
prevCumReward = 0;
cumIntrReward = 0;
prevCumIntrReward = 0;
maxFeatVectorNorm = 1;
pathToSaveLearnedWeights = param->outputPath;
for(int i = 0; i < (ramFeatures.getNumberOfFeatures() - 1) * 2; i++){
transitions.push_back(0);
}
actions = ale.getLegalActionSet();
numOptions = param->numOptions;
numBasicActions = actions.size();
numTotalActions = numBasicActions + numOptions;
//Reading file containing the vector that describes the reward for the option learning
//The first X positions encode the transition 0->1 and the other X encode 1->0.
pathToRewardDescription = param->eigVectorPath;
std::ifstream infile1(pathToRewardDescription.c_str());
float value;
while(infile1 >> value){
eigVector.push_back(value);
}
pathToStatsDescription = param->statEigVectorPath;
std::ifstream infile2((pathToStatsDescription + "_mean.out").c_str());
while(infile2 >> value){
mean.push_back(value);
}
std::ifstream infile3((pathToStatsDescription + "_std.out").c_str());
while(infile3 >> value){
std.push_back(value);
}
numFeatures = bproFeatures.getNumberOfFeatures();
for(int i = 0; i < numTotalActions; i++){
//Initialize Q;
Q.push_back(0);
Qnext.push_back(0);
//Initialize e:
e.push_back(vector<float>(numFeatures, 0.0));
w.push_back(vector<float>(numFeatures, 0.0));
nonZeroElig.push_back(vector<int>());
}
}
int main(int argc, char** argv) {
if (argc < 2) {
std::cerr << "Usage: " << argv[0] << " rom_file" << std::endl;
return 1;
}
ALEInterface ale;
// Get & Set the desired settings
ale.setInt("random_seed", 123);
//The default is already 0.25, this is just an example
ale.setFloat("repeat_action_probability", 0.25);
#ifdef __USE_SDL
ale.setBool("display_screen", true);
ale.setBool("sound", true);
#endif
// Load the ROM file. (Also resets the system for new settings to
// take effect.)
ale.loadROM(argv[1]);
// Get the vector of legal actions
ActionVect legal_actions = ale.getLegalActionSet();
// Play 10 episodes
for (int episode=0; episode<10; episode++) {
float totalReward = 0;
while (!ale.game_over()) {
Action a = legal_actions[rand() % legal_actions.size()];
// Apply the action and get the resulting reward
float reward = ale.act(a);
totalReward += reward;
if (reward != 0){
cout<<"totalReward: "<<totalReward<<endl;
}
}
cout << "Episode " << episode << " ended with score: " << totalReward << endl;
ale.reset_game();
}
return 0;
}
RLLearner::RLLearner(ALEInterface& ale, Parameters *param){
frame = 0;
randomActionTaken = 0;
gamma = param->getGamma();
epsilon = param->getEpsilon();
toUseOnlyRewardSign = param->getUseRewardSign();
numStepsPerAction = param->getNumStepsPerAction();
toBeOptimistic = param->getOptimisticInitialization();
episodeLength = param->getEpisodeLength();
numEpisodesEval = param->getNumEpisodesEval();
numEpisodesLearn = param->getNumEpisodesLearn();
//Get the number of effective actions:
if(param->isMinimalAction()){
actions = ale.getMinimalActionSet();
}
else{
actions = ale.getLegalActionSet();
}
numActions = actions.size();
}