CTTableDiff<COUNT,PROB>* model1::one_step_em(int it, bool seedModel1,
Dictionary& dictionary, bool useDict)
{
CTTableDiff<COUNT,PROB> *diff = new CTTableDiff<COUNT,PROB>();
double minErrors=1.0;
string modelName="Model1",shortModelName="1";
time_t st, it_st, fn;
string tfile, number, alignfile, test_alignfile;
bool dump_files = false ;
st = time(NULL);
sHandler1.rewind();
cout << "==========================================================\n";
cout << modelName << " Training Started at: "<< my_ctime(&st) << "\n";
it_st = time(NULL);
cout << "-----------\n" << modelName << ": Iteration " << it << '\n';
dump_files = (Model1_Dump_Freq != 0) && ((it % Model1_Dump_Freq) == 0) && !NODUMPS ;
number = "";
int n = it;
do {
number.insert((size_t)0, 1, (char)(n % 10 + '0'));
} while((n /= 10) > 0);
tfile = Prefix + ".t" + shortModelName + "." + number ;
alignfile = Prefix + ".A1" ;
test_alignfile = Prefix +".tst.A" + shortModelName + "." + number ;
initAL();
em_loop_1(diff,it,perp, sHandler1, seedModel1,
dump_files, alignfile.c_str(), dictionary, useDict, trainViterbiPerp);
//if (testPerp && testHandler) // calculate test perplexity
// em_loop(it,*testPerp, *testHandler, seedModel1, dump_files, test_alignfile.c_str(), dictionary, useDict, *testViterbiPerp, true);
if( errorsAL()<minErrors ) minErrors=errorsAL();
fn = time(NULL) ;
cout << "Partial " << modelName << " Training took: " << difftime(fn, it_st) << " seconds\n";
return diff;
}
void Game::initialise (int argc, char *argv[]) {
// Initialise libraries
screen = NULL;
errorLoadingLibraries = false;
running = false;
paused = false;
initTime = SDL_GetTicks ();
try {
initGUI ();
initGL ();
resizeWindow (screen->w, screen->h);
initAL ();
initDecoder ();
} catch (const char *message) {
errorLoadingLibraries = true;
std::cerr << message << std::endl;
}
}
void loadResources(Resources *resources)
{
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR_MIPMAP_LINEAR);
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_REPEAT);
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_REPEAT);
glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_MODULATE);
glEnable(GL_TEXTURE_2D);
initAL(resources);
std::string font_path;
search_path("FreeMonoBold.ttf", font_path);
font = new FTGLBitmapFont(font_path.c_str());
Fontbig = new FTGLBitmapFont(font_path.c_str());
if (font->Error()) {
DEBUG_ERROR("error on font");
}
font->FaceSize(18);
if (Fontbig->Error()) {
DEBUG_ERROR("error on font");
}
Fontbig->FaceSize(110);
restartButton = new Button(10, 10, 100, 50, Resources_Manager.texture("restart_button.tga"), restartMatch);
newMazeButton = new Button(10, 10, 100, 50, Resources_Manager.texture("newmaze_button.tga"), newMatch);
toggleCameraButton = new Button(10, 10, 100, 50, Resources_Manager.texture("togglecamera_button.tga"), toggleCameraMode);
Widget_List.push_back(restartButton);
Widget_List.push_back(newMazeButton);
Widget_List.push_back(toggleCameraButton);
skybox.setTexture(Resources_Manager.texture("checkerboard.tga"));
skybox.build(Config_Info.matrixDimensions.x*Tilesize*20.0f);
}
int model1::em_with_tricks(int noIterations, /*Perplexity& perp, sentenceHandler& sHandler1, */
bool seedModel1, Dictionary& dictionary, bool useDict, Word2Vec& word2vec, bool useWord2Vec /*Perplexity* testPerp, sentenceHandler* testHandler,
Perplexity& trainViterbiPerp, Perplexity* testViterbiPerp */ )
{
double minErrors=1.0;int minIter=0;
string modelName="Model1",shortModelName="1";
time_t st, it_st, fn, it_fn;
string tfile, number, alignfile, test_alignfile;
int pair_no;
bool dump_files = false ;
st = time(NULL);
if(useWord2Vec && word2vec.Method == 2){
word2vec.computeSimilarities(sHandler1);
}
sHandler1.rewind();
cout << "==========================================================\n";
cout << modelName << " Training Started at: "<< ctime(&st) << "\n";
cout << "useDict: " << useDict << endl;
for(int it = 1; it <= noIterations; it++){
pair_no = 0 ;
it_st = time(NULL);
cout << "-----------\n" << modelName << ": Iteration " << it << '\n';
dump_files = (Model1_Dump_Freq != 0) && ((it % Model1_Dump_Freq) == 0) && !NODUMPS ;
number = "";
int n = it;
do{
number.insert((size_t)0, 1, (char)(n % 10 + '0'));
} while((n /= 10) > 0);
tfile = Prefix + ".t" + shortModelName + "." + number ;
alignfile = Prefix + ".A" + shortModelName + "." + number ;
test_alignfile = Prefix +".tst.A" + shortModelName + "." + number ;
initAL();
em_loop(it,perp, sHandler1, seedModel1, dump_files, alignfile.c_str(), dictionary, useDict, word2vec, useWord2Vec, trainViterbiPerp);
if (testPerp && testHandler) // calculate test perplexity
em_loop(it,*testPerp, *testHandler, seedModel1, dump_files, test_alignfile.c_str(), dictionary, useDict, word2vec, useWord2Vec, *testViterbiPerp, true);
if( errorsAL()<minErrors )
{
minErrors=errorsAL();
minIter=it;
}
if (dump_files){
if( OutputInAachenFormat==1 )
tTable.printCountTable(tfile.c_str(),Elist.getVocabList(),Flist.getVocabList(),1);
}
tTable.normalizeTable(Elist, Flist);
cout << modelName << ": ("<<it<<") TRAIN CROSS-ENTROPY " << perp.cross_entropy()
<< " PERPLEXITY " << perp.perplexity() << '\n';
if (testPerp && testHandler)
cout << modelName << ": ("<<it<<") TEST CROSS-ENTROPY " << (*testPerp).cross_entropy()
<< " PERPLEXITY " << (*testPerp).perplexity()
<< '\n';
cout << modelName << ": ("<<it<<") VITERBI TRAIN CROSS-ENTROPY " << trainViterbiPerp.cross_entropy()
<< " PERPLEXITY " << trainViterbiPerp.perplexity() << '\n';
if (testPerp && testHandler)
cout << modelName << ": ("<<it<<") VITERBI TEST CROSS-ENTROPY " << (*testViterbiPerp).cross_entropy()
<< " PERPLEXITY " << (*testViterbiPerp).perplexity()
<< '\n';
if (dump_files){
if( OutputInAachenFormat==0 )
tTable.printProbTable(tfile.c_str(),Elist.getVocabList(),Flist.getVocabList(),OutputInAachenFormat);
}
it_fn = time(NULL);
cout << "Model 1 Iteration: " << it<< " took: " << difftime(it_fn, it_st) << " seconds\n";
}
fn = time(NULL) ;
cout << "Entire " << modelName << " Training took: " << difftime(fn, st) << " seconds\n";
return minIter;
}
int model2::em_with_tricks(int noIterations)
{
double minErrors=1.0;int minIter=0;
string modelName="Model2",shortModelName="2";
time_t it_st, st, it_fn, fn;
string tfile, afile, number, alignfile, test_alignfile;
int pair_no = 0;
bool dump_files = false ;
ofstream of2 ;
st = time(NULL) ;
sHandler1.rewind();
cout << "\n==========================================================\n";
cout << modelName << " Training Started at: " << ctime(&st) << " iter: " << noIterations << "\n";
for(int it=1; it <= noIterations ; it++){
pair_no = 0;
it_st = time(NULL) ;
cout << endl << "-----------\n" << modelName << ": Iteration " << it << '\n';
dump_files = (Model2_Dump_Freq != 0) && ((it % Model2_Dump_Freq) == 0) && !NODUMPS;
number = "";
int n = it;
do{
number.insert((size_t)0, 1, (char)(n % 10 + '0'));
} while((n /= 10) > 0);
tfile = Prefix + ".t" + shortModelName + "." + number ;
afile = Prefix + ".a" + shortModelName + "." + number ;
alignfile = Prefix + ".A" + shortModelName + "." + number ;
test_alignfile = Prefix + ".tst.A" + shortModelName + "." + number ;
aCountTable.clear();
initAL();
em_loop(perp, sHandler1, dump_files, alignfile.c_str(), trainViterbiPerp, false);
if( errorsAL()<minErrors )
{
minErrors=errorsAL();
minIter=it;
}
if (testPerp && testHandler)
em_loop(*testPerp, *testHandler, dump_files, test_alignfile.c_str(), *testViterbiPerp, true);
if (dump_files&&OutputInAachenFormat==1)
tTable.printCountTable(tfile.c_str(),Elist.getVocabList(),Flist.getVocabList(),1);
tTable.normalizeTable(Elist, Flist);
aCountTable.normalize(aTable);
cout << modelName << ": ("<<it<<") TRAIN CROSS-ENTROPY " << perp.cross_entropy()
<< " PERPLEXITY " << perp.perplexity() << '\n';
if (testPerp && testHandler)
cout << modelName << ": ("<<it<<") TEST CROSS-ENTROPY " << (*testPerp).cross_entropy()
<< " PERPLEXITY " << (*testPerp).perplexity()
<< '\n';
cout << modelName << ": ("<<it<<") VITERBI TRAIN CROSS-ENTROPY " << trainViterbiPerp.cross_entropy()
<< " PERPLEXITY " << trainViterbiPerp.perplexity() << '\n';
if (testPerp && testHandler)
cout << modelName << ": ("<<it<<") VITERBI TEST CROSS-ENTROPY " << testViterbiPerp->cross_entropy()
<< " PERPLEXITY " << testViterbiPerp->perplexity()
<< '\n';
if (dump_files)
{
if(OutputInAachenFormat==0)
tTable.printProbTable(tfile.c_str(),Elist.getVocabList(),Flist.getVocabList(),OutputInAachenFormat);
aCountTable.printTable(afile.c_str());
}
it_fn = time(NULL) ;
cout << modelName << " Iteration: " << it<< " took: " << difftime(it_fn, it_st) << " seconds\n";
} // end of iterations
aCountTable.clear();
fn = time(NULL) ;
cout << endl << "Entire " << modelName << " Training took: " << difftime(fn, st) << " seconds\n";
// cout << "tTable contains " << tTable.getHash().bucket_count()
// << " buckets and " << tTable.getHash().size() << " entries." ;
cout << "==========================================================\n";
return minIter;
}
int model1::em_with_tricks(int noIterations, bool seedModel1, Dictionary& dictionary, bool useDict)
{
double minErrors=1.0;int minIter=0;
string modelName="Model1",shortModelName="1";
time_t st, it_st, fn, it_fn;
string tfile, number, alignfile, test_alignfile;
int pair_no;
bool dump_files = false ;
st = time(NULL);
sHandler1.rewind();
cout << "==========================================================\n";
cout << modelName << " Training Started at: "<< ctime(&st) << "\n";
for(int it = 1; it <= noIterations; it++){
pair_no = 0 ;
srcHits_.clear();
it_st = time(NULL);
cout << "-----------\n" << modelName << ": Iteration " << it << '\n';
dump_files = (Model1_Dump_Freq != 0) && ((it % Model1_Dump_Freq) == 0) && !NODUMPS ;
number = "";
int n = it;
do{
number.insert((size_t)0, 1, (char)(n % 10 + '0'));
} while((n /= 10) > 0);
tfile = Prefix + ".t" + shortModelName + "." + number ;
alignfile = Prefix + ".A" + shortModelName + "." + number ;
test_alignfile = Prefix +".tst.A" + shortModelName + "." + number ;
initAL();
em_loop(it,perp, sHandler1, seedModel1, dump_files, alignfile.c_str(), dictionary, useDict, trainViterbiPerp);
if (testPerp && testHandler) // calculate test perplexity
em_loop(it,*testPerp, *testHandler, seedModel1, dump_files, test_alignfile.c_str(), dictionary, useDict, *testViterbiPerp, true);
if( errorsAL()<minErrors ) {
minErrors=errorsAL();
minIter=it;
}
/*if (dump_files){
if( OutputInAachenFormat==1 )
tTable.printCountTable(tfile.c_str(),Elist.getVocabList(),Flist.getVocabList(),1);
}*/
if(step_k != 0) tTable.interpolateStepCounts(); // interpolate stepcounts and full counts here
tTable.normalizeTable(Elist, Flist, &srcHits_);
cout << modelName << ": ("<<it<<") TRAIN CROSS-ENTROPY " << perp.cross_entropy()
<< " PERPLEXITY " << perp.perplexity() << '\n';
if (testPerp && testHandler)
cout << modelName << ": ("<<it<<") TEST CROSS-ENTROPY " << (*testPerp).cross_entropy()
<< " PERPLEXITY " << (*testPerp).perplexity()
<< '\n';
cout << modelName << ": ("<<it<<") VITERBI TRAIN CROSS-ENTROPY " << trainViterbiPerp.cross_entropy()
<< " PERPLEXITY " << trainViterbiPerp.perplexity() << '\n';
if (testPerp && testHandler)
cout << modelName << ": ("<<it<<") VITERBI TEST CROSS-ENTROPY " << (*testViterbiPerp).cross_entropy()
<< " PERPLEXITY " << (*testViterbiPerp).perplexity()
<< '\n';
if (dump_files && (!run_giza_server)){
tTable.printProbTable(tfile.c_str(),Elist.getVocabList(),Flist.getVocabList(),OutputInAachenFormat);
}
it_fn = time(NULL);
cout << "Model 1 Iteration: " << it<< " took: " << difftime(it_fn, it_st) << " seconds\n";
} // end noIterations
fn = time(NULL) ;
cout << "Entire " << modelName << " Training took: " << difftime(fn, st) << " seconds\n";
return minIter;
}
//@ASHISH VASWANI: A FUNCTION THAT RUNS A SINGLE ITERATION OF MODEL1. WILL BE NEEDED TO RUN E-F and
//F-E SEPARATELY
int model1::em_with_tricks_e_step(int iterationNumber, /*Perplexity& perp, sentenceHandler& sHandler1, */
bool seedModel1,
Dictionary& dictionary,
bool useDict,
string model_prefix /*Perplexity* testPerp, sentenceHandler* testHandler,
Perplexity& trainViterbiPerp, Perplexity* testViterbiPerp */ )
{
double minErrors=1.0;int minIter=0;
string modelName="Model1",shortModelName="1";
time_t st, it_st, fn, it_fn;
string tfile, number, alignfile, test_alignfile;
int pair_no;
bool dump_files = false ;
st = time(NULL);
sHandler1.rewind();
cout << "==========================================================\n";
cout << modelName << " Training for iteration number "<<iterationNumber<<" Started at: "<< ctime(&st) << "\n";
//for(int it = 1; it <= noIterations; it++){
int it = iterationNumber;
pair_no = 0 ;
it_st = time(NULL);
cout << "-----------\n" << modelName << ": Iteration " << it << '\n';
dump_files = (Model1_Dump_Freq != 0) && ((it % Model1_Dump_Freq) == 0) && !NODUMPS ;
number = "";
int n = it;
do{
number.insert((size_t)0, 1, (char)(n % 10 + '0'));
} while((n /= 10) > 0);
tfile = Prefix + ".t" + model_prefix + "." + shortModelName + "." + number ;
alignfile = Prefix + ".A" + model_prefix + "." + shortModelName + "." + number ;
test_alignfile = Prefix +".tst.A" + model_prefix + "." + shortModelName + "." + number ;
initAL();
em_loop(it,perp, sHandler1, seedModel1, dump_files, alignfile.c_str(), dictionary, useDict, trainViterbiPerp);
if (testPerp && testHandler) // calculate test perplexity
em_loop(it,*testPerp, *testHandler, seedModel1, dump_files, test_alignfile.c_str(), dictionary, useDict, *testViterbiPerp, true);
if( errorsAL()<minErrors )
{
minErrors=errorsAL();
minIter=it;
}
if (dump_files){
if( OutputInAachenFormat==1 )
tTable.printCountTable(tfile.c_str(),Elist.getVocabList(),Flist.getVocabList(),1);
}
/*
// accumulate expected counts
vector<vector<COUNT> > expCntsVec,probsVec;
vector<float> rowwiseExpCntsSum;
cout<<" Accumulating expected counts "<<endl;
tTable.getCounts(&expCntsVec,&rowwiseExpCntsSum);
//printCounts(expCntsVec);
//getchar();
cout<<" Accumulating probabilities"<<endl;
tTable.getProbs(&probsVec);
//printCounts(probsVec);
//getchar();
cout<<" Normalizing t table"<<endl;
//tTable.normalizeTable(Elist, Flist);
*/
/*
cout << modelName << ": ("<<it<<") TRAIN CROSS-ENTROPY " << perp.cross_entropy()
<< " PERPLEXITY " << perp.perplexity() << '\n';
if (testPerp && testHandler)
cout << modelName << ": ("<<it<<") TEST CROSS-ENTROPY " << (*testPerp).cross_entropy()
<< " PERPLEXITY " << (*testPerp).perplexity()
<< '\n';
cout << modelName << ": ("<<it<<") VITERBI TRAIN CROSS-ENTROPY " << trainViterbiPerp.cross_entropy()
<< " PERPLEXITY " << trainViterbiPerp.perplexity() << '\n';
if (testPerp && testHandler)
cout << modelName << ": ("<<it<<") VITERBI TEST CROSS-ENTROPY " << (*testViterbiPerp).cross_entropy()
<< " PERPLEXITY " << (*testViterbiPerp).perplexity()
<< '\n';
if (dump_files){
if( OutputInAachenFormat==0 )
tTable.printProbTable(tfile.c_str(),Elist.getVocabList(),Flist.getVocabList(),OutputInAachenFormat);
}
*/
it_fn = time(NULL);
cout << "Model 1 Iteration: " << it<< " took: " << difftime(it_fn, it_st) << " seconds\n";
//}
fn = time(NULL) ;
cout << "Entire " << modelName << " Training of iteration number "<<iterationNumber<<" took: " << difftime(fn, st) << " seconds\n";
return minIter;
}
View::View( QWidget * parent ) : QGraphicsView( parent )
{
initGL();
initAL();
initScene();
}
int model1::em_with_tricks(int noIterations, /*Perplexity& perp, sentenceHandler& sHandler1, */
bool seedModel1, Dictionary& dictionary, bool useDict /*Perplexity* testPerp, sentenceHandler* testHandler,
Perplexity& trainViterbiPerp, Perplexity* testViterbiPerp */
, bool dumpCount , const char* dumpCountName, bool useString) // If specified, then will dump files before last iteration
{
double minErrors=1.0;
int minIter=0;
string modelName="Model1",shortModelName="1";
time_t st, it_st, fn, it_fn;
string tfile, number, alignfile, test_alignfile;
bool dump_files = false ;
st = time(NULL);
sHandler1.rewind();
cout << "==========================================================\n";
cout << modelName << " Training Started at: "<< my_ctime(&st) << "\n";
for(int it = 1; it <= noIterations; it++) {
it_st = time(NULL);
cout << "-----------\n" << modelName << ": Iteration " << it << '\n';
dump_files = (Model1_Dump_Freq != 0) && ((it % Model1_Dump_Freq) == 0 || it == noIterations) && !NODUMPS ;
//dump_files = true;
number = "";
int n = it;
do {
number.insert((size_t)0, 1, (char)(n % 10 + '0'));
} while((n /= 10) > 0);
tfile = Prefix + ".t" + shortModelName + "." + number ;
alignfile = Prefix + ".A" + shortModelName + "." + number+".part0" ;
test_alignfile = Prefix +".tst.A" + shortModelName + "." + number ;
initAL();
threadID = 0;
int th;
vector<em_loop_t> ths;
ths.resize(NCPUS);
sHandler1.rewind();
for (th=1; th<NCPUS; th++) {
ths[th].m1=this;
ths[th].it = it;
ths[th].nthread = th;
ths[th].dict = & dictionary;
ths[th].useDict = useDict;
ths[th].result = 0;
ths[th].valid = pthread_create(&(ths[th].thread),NULL,exe_emloop,&(ths[th]));
if(ths[th].valid) {
cerr << "Error starting thread " << th << endl;
}
}
em_loop(it,perp, sHandler1, seedModel1, dump_files, alignfile.c_str(), dictionary, useDict, trainViterbiPerp);
perp.record("Model1");
trainViterbiPerp.record("Model1");
errorReportAL(cout, "IBM-1");
cerr << "Main thread done, waiting" << endl;;
for (th=1; th<NCPUS; th++) {
pthread_join((ths[th].thread),NULL);
cerr << "Thread " << th << "done" << endl;
}
if (testPerp && testHandler) // calculate test perplexity
em_loop(it,*testPerp, *testHandler, seedModel1, dump_files, test_alignfile.c_str(), dictionary, useDict, *testViterbiPerp, true);
if( errorsAL()<minErrors ) {
minErrors=errorsAL();
minIter=it;
}
//if (dump_files){
// if( OutputInAachenFormat==1 )
// tTable.printCountTable(tfile.c_str(),Elist.getVocabList(),Flist.getVocabList(),1);
//}
cerr << "Normalizing T " << endl;
/**
If asked for dumping count table, just dump it.
*/
if(dumpCount && it == noIterations) {
string realTableName = dumpCountName;
realTableName += ".t.count";
tTable.printCountTable(realTableName.c_str(),Elist.getVocabList(),Flist.getVocabList(),useString);
}
tTable.normalizeTable(Elist, Flist);
//cout << tTable.getProb(2,2) << endl;
cerr << " DONE Normalizing " << endl;
cout << modelName << ": ("<<it<<") TRAIN CROSS-ENTROPY " << perp.cross_entropy()
<< " PERPLEXITY " << perp.perplexity() << '\n';
if (testPerp && testHandler)
cout << modelName << ": ("<<it<<") TEST CROSS-ENTROPY " << (*testPerp).cross_entropy()
<< " PERPLEXITY " << (*testPerp).perplexity()
<< '\n';
cout << modelName << ": ("<<it<<") VITERBI TRAIN CROSS-ENTROPY " << trainViterbiPerp.cross_entropy()
<< " PERPLEXITY " << trainViterbiPerp.perplexity() << '\n';
if (testPerp && testHandler)
cout << modelName << ": ("<<
it<<") VITERBI TEST CROSS-ENTROPY "
<< (*testViterbiPerp).cross_entropy()
<< " PERPLEXITY " << (*testViterbiPerp).perplexity()
<< '\n';
if (dump_files) {
if( OutputInAachenFormat==0 )
tTable.printProbTable(tfile.c_str(),Elist.getVocabList(),
Flist.getVocabList(),OutputInAachenFormat);
}
it_fn = time(NULL);
cout << "Model 1 Iteration: " << it<< " took: " << difftime(it_fn, it_st) << " seconds\n";
}
fn = time(NULL) ;
cout << "Entire " << modelName << " Training took: " << difftime(fn, st) << " seconds\n";
return minIter;
}
