void testhmm(string race, int index)
{
cout << "Testing with replay " << index << endl;
Hmm hmm;
hmm.loadFromRace(race);
vector<unsigned long>* replay = loadReplayData(race, index);
cout << "Replay length: " << replay->size() << endl << "Rep:";
for (vector<unsigned long>::iterator it = replay->begin(); it != replay->end(); it++) {
cout.width(3); cout << (*it);
}
for (unsigned int i = 0; i < replay->size(); i++) {
unsigned int missed = 0, j;
cout << endl << "Pre:";
for (j = 0; j < i; j++) {
cout.width(3);
cout << replay->at(j);
hmm.observe(replay->at(j));
}
vector<unsigned long> *seq = hmm.predictMaxSeq(replay->size() - i);
for (unsigned int k = j; k < replay->size(); k++) {
cout.width(3);
unsigned long prediction = seq->at(k - j);
if (prediction != replay->at(k)){
missed++;
cout << red << prediction;
}
else {
cout << green << prediction;
}
}
cout << white << " (" << (1.0 - missed / (double)(replay->size() - i)) << ")" << endl;
hmm.reset();
delete seq;
}
}
void testBuildingStats() {
Hmm hmm;
hmm.loadFromRace("P");
BuildingStats stats;
stats.readStatsFile("Z/stats.csv");
stats.readOurStatsFile("P/stats.csv");
stats.readRepliesFile("Z/replies.csv");
int rep_state = stats.getReplyState(11);
set<string> search;
search.insert("Reaver");
search.insert("Observer");
int state = stats.getClosestState(search);
int state2 = stats.getClosestState2(search);
cout << state << " " << state2 << endl;
set<string>* unitTypes = stats.decodeState(state);
for (set<string>::iterator it = unitTypes->begin(); it != unitTypes->end(); it++) {
cout << *it << ",";
}
cout << endl << endl;;
set<string>* unitTypes2 = stats.decodeState(state2);
for (set<string>::iterator it = unitTypes2->begin(); it != unitTypes2->end(); it++) {
cout << *it << ",";
}
cout << endl;
}
void testdata(string race, int numLines, int *trials, int *correct, int advance) {
Hmm hmm;
hmm.loadFromRace(race);
string filename = race + "/data.csv";
ifstream infile(filename.c_str());
string line;
for (int z = 0; z < numLines; z++) {
vector<unsigned long>* replay = loadReplayData(race, z);
for (unsigned int i = 0; i < replay->size(); i++) {
unsigned int missed = 0, j;
for (j = 0; j < i; j++) {
hmm.observe(replay->at(j));
}
vector<unsigned long> *seq = hmm.predictMaxSeq(advance);
for (unsigned int k = j - 1 + advance; k < replay->size() && k < j + advance; k++) {
unsigned long prediction = seq->at(k - j);
if (prediction == replay->at(k)){
correct[k]++;
}
trials[k]++;
}
hmm.reset();
delete seq;
}
cerr << "Ran " << z << " from " << numLines << endl;
}
}
// construit les matrices emit et trans à l'aide de l'algorithme Baum-Welch
void learn(const std::string& name, unsigned int maxIterations)
{
Hmm hmm;
std::string output = (boost::format("seq-%1%-result") %
name).str();
ifstream istrm((boost::format("seq-%1%.input") % name).str().c_str());
std:: vector < std::vector < unsigned long >* > trainingSequences;
hmm.loadProbs((boost::format("seq-%1%-init") % name).str().c_str());
hmm.readSeqs(istrm, trainingSequences);
hmm.baumWelch(trainingSequences, maxIterations);
hmm.saveProbs(output.c_str());
}
// génére les dates des événements de sortie en fonction du HMM sur
// une période [0, max]
// le paramètre events est la liste des dates d'arrivée des événements d'entrée
Dates generate_dates(double max, const Dates& events, Hmm& hmm)
{
Dates dates;
double t = 0;
unsigned long state = hmm.getInitState();
Dates::const_iterator ite = events.begin();
while (t < max) {
unsigned long obs;
if (hmm.genSeq(state, obs)) {
std::string obsStr = hmm.getStr(obs);
int type;
std::vector < std::string > obsV;
boost::split(obsV, obsStr, boost::is_any_of("|"));
type = atoi(obsV[1].c_str());
if (type != IN) {
if (type != END) {
// ce n'est pas un état infini
if (type != INF) {
double value = atof(obsV[0].c_str());
if (dates.empty()) {
dates.push_back(value);
} else {
dates.push_back(dates.back() + value);
}
t = dates.back();
// détermine la prochaine date d'arrivée d'un
// événement d'entrée
while (ite != events.end() and t >= *ite) {
++ite;
}
} else {
// c'est un état infini
if (ite != events.end()) {
t = *ite;
++ite;
}
}
} else {
state = hmm.getInitState();
}
} else {
state = hmm.getInitState();
}
}
}
return dates;
}
int main(int argc, char* argv[])
{
Hmm hmm;
if (argc<3) {
cerr << "USAGE: trainhmm INIT-HMM RESULT-HMM DATA [MAX-ITERATIONS]" << endl;
exit(1);
}
hmm.loadProbs(argv[1]);
const char* output = argv[2];
ifstream istrm(argv[3]);
int maxIterations = 10;
if (argc>4)
maxIterations = atoi(argv[4]);
vector<vector<unsigned long>*> trainingSequences;
hmm.readSeqs(istrm, trainingSequences);
hmm.baumWelch(trainingSequences, maxIterations);
hmm.saveProbs(output);
}
void trainhmm(string race, int numStates, int maxIterations)
{
Hmm hmm;
hmm.makeEmitAndTransFiles(race, numStates);
hmm.loadFromRace(race);
ifstream istrm((race + "/data.csv").c_str());
vector<vector<unsigned long>*> trainingSequences;
hmm.readSeqs(istrm, trainingSequences);
hmm.baumWelch(trainingSequences, maxIterations);
hmm.saveProbs(race + "/hmm");
}
virtual std::string say_hmm()
{
return hmm_.say();
}