// Returns the results into ptRes
void MDDAGClassifier::computeResults(InputData* pData, vector<BaseLearner*>& weakHypotheses,
vector< ExampleResults* >& results, int numIterations)
{
assert( !weakHypotheses.empty() );
const int numClasses = pData->getNumClasses();
const int numExamples = pData->getNumExamples();
// Initialize the output info
OutputInfo* pOutInfo = NULL;
if ( !_outputInfoFile.empty() )
{
if ( _args.getNumValues("outputinfo") > 1 )
{
pOutInfo = new OutputInfo(_args);;
}
else
{
pOutInfo = new OutputInfo(_outputInfoFile, "e01hamauc", false);
}
}
// Creating the results structures. See file Structures.h for the
// PointResults structure
results.clear();
results.reserve(numExamples);
for (int i = 0; i < numExamples; ++i)
results.push_back( new ExampleResults(i, numClasses) );
// iterator over all the weak hypotheses
vector<BaseLearner*>::const_iterator whyIt;
int t;
if ( pOutInfo )
{
pOutInfo->initialize( pData );
pOutInfo->outputHeader(pData->getClassMap(),
true, // output iterations
false, // output time
true // endline
);
}
// for every feature: 1..T
for (whyIt = weakHypotheses.begin(), t = 0;
whyIt != weakHypotheses.end() && t < numIterations; ++whyIt, ++t)
{
BaseLearner* currWeakHyp = *whyIt;
AlphaReal alpha = currWeakHyp->getAlpha();
// for every point
for (int i = 0; i < numExamples; ++i)
{
// a reference for clarity and speed
vector<AlphaReal>& currVotesVector = results[i]->getVotesVector();
// for every class
for (int l = 0; l < numClasses; ++l)
currVotesVector[l] += alpha * currWeakHyp->classify(pData, i, l);
}
// if needed output the step-by-step information
if ( pOutInfo )
{
pOutInfo->outputIteration(t);
// pOutInfo->outputError(pData, currWeakHyp);
// pOutInfo->outTPRFPR(pData);
//pOutInfo->outputBalancedError(pData, currWeakHyp);
// if ( ( t % 1 ) == 0 ) {
// pOutInfo->outputROC(pData);
// }
pOutInfo->outputCustom(pData, currWeakHyp);
// Margins and edge requires an update of the weight,
// therefore I keep them out for the moment
//outInfo.outputMargins(pData, currWeakHyp);
//outInfo.outputEdge(pData, currWeakHyp);
pOutInfo->endLine();
}
}
if (pOutInfo)
delete pOutInfo;
}
void FilterBoostLearner::run(const nor_utils::Args& args)
{
// load the arguments
this->getArgs(args);
time_t startTime, currentTime;
time(&startTime);
// get the registered weak learner (type from name)
BaseLearner* pWeakHypothesisSource =
BaseLearner::RegisteredLearners().getLearner(_baseLearnerName);
// initialize learning options; normally it's done in the strong loop
// also, here we do it for Product learners, so input data can be created
pWeakHypothesisSource->initLearningOptions(args);
BaseLearner* pConstantWeakHypothesisSource =
BaseLearner::RegisteredLearners().getLearner("ConstantLearner");
// get the training input data, and load it
InputData* pTrainingData = pWeakHypothesisSource->createInputData();
pTrainingData->initOptions(args);
pTrainingData->load(_trainFileName, IT_TRAIN, _verbose);
const int numClasses = pTrainingData->getNumClasses();
const int numExamples = pTrainingData->getNumExamples();
//initialize the margins variable
_margins.resize( numExamples );
for( int i=0; i<numExamples; i++ )
{
_margins[i].resize( numClasses );
fill( _margins[i].begin(), _margins[i].end(), 0.0 );
}
// get the testing input data, and load it
InputData* pTestData = NULL;
if ( !_testFileName.empty() )
{
pTestData = pWeakHypothesisSource->createInputData();
pTestData->initOptions(args);
pTestData->load(_testFileName, IT_TEST, _verbose);
}
// The output information object
OutputInfo* pOutInfo = NULL;
if ( !_outputInfoFile.empty() )
{
// Baseline: constant classifier - goes into 0th iteration
BaseLearner* pConstantWeakHypothesis = pConstantWeakHypothesisSource->create() ;
pConstantWeakHypothesis->initLearningOptions(args);
pConstantWeakHypothesis->setTrainingData(pTrainingData);
AlphaReal constantEnergy = pConstantWeakHypothesis->run();
pOutInfo = new OutputInfo(args);
pOutInfo->initialize(pTrainingData);
updateMargins( pTrainingData, pConstantWeakHypothesis );
if (pTestData)
pOutInfo->initialize(pTestData);
pOutInfo->outputHeader(pTrainingData->getClassMap() );
pOutInfo->outputIteration(-1);
pOutInfo->outputCustom(pTrainingData, pConstantWeakHypothesis);
if (pTestData)
{
pOutInfo->separator();
pOutInfo->outputCustom(pTestData, pConstantWeakHypothesis);
}
pOutInfo->outputCurrentTime();
pOutInfo->endLine();
pOutInfo->initialize(pTrainingData);
if (pTestData)
pOutInfo->initialize(pTestData);
}
// reload the previously found weak learners if -resume is set.
// otherwise just return 0
int startingIteration = resumeWeakLearners(pTrainingData);
Serialization ss(_shypFileName, _isShypCompressed );
ss.writeHeader(_baseLearnerName); // this must go after resumeProcess has been called
// perform the resuming if necessary. If not it will just return
resumeProcess(ss, pTrainingData, pTestData, pOutInfo);
if (_verbose == 1)
cout << "Learning in progress..." << endl;
///////////////////////////////////////////////////////////////////////
// Starting the AdaBoost main loop
///////////////////////////////////////////////////////////////////////
for (int t = startingIteration; t < _numIterations; ++t)
{
if (_verbose > 1)
cout << "------- WORKING ON ITERATION " << (t+1) << " -------" << endl;
// create the weak learner
BaseLearner* pWeakHypothesis;
BaseLearner* pConstantWeakHypothesis;
pWeakHypothesis = pWeakHypothesisSource->create();
pWeakHypothesis->initLearningOptions(args);
//pTrainingData->clearIndexSet();
pWeakHypothesis->setTrainingData(pTrainingData);
AlphaReal edge, energy=0.0;
// create the constant learner
pConstantWeakHypothesis = pConstantWeakHypothesisSource->create() ;
pConstantWeakHypothesis->initLearningOptions(args);
pConstantWeakHypothesis->setTrainingData(pTrainingData);
AlphaReal constantEdge = -numeric_limits<AlphaReal>::max();
int currentNumberOfUsedData = static_cast<int>(_Cn * log(t+3.0));
if ( _onlineWeakLearning )
{
//check whether the weak learner is a ScalarLeaerner
try {
StochasticLearner* pStochasticLearner = dynamic_cast<StochasticLearner*>(pWeakHypothesis);
StochasticLearner* pStochasticConstantWeakHypothesis = dynamic_cast<StochasticLearner*> (pConstantWeakHypothesis);
pStochasticLearner->initLearning();
pStochasticConstantWeakHypothesis->initLearning();
if (_verbose>1)
cout << "Number of random instances: \t" << currentNumberOfUsedData << endl;
// set the weights
setWeightToMargins(pTrainingData);
//learning
for (int i=0; i<currentNumberOfUsedData; ++i )
{
int randomIndex = (rand() % pTrainingData->getNumExamples());
//int randomIndex = getRandomIndex();
pStochasticLearner->update(randomIndex);
pStochasticConstantWeakHypothesis->update(randomIndex);
}
pStochasticLearner->finishLearning();
pStochasticConstantWeakHypothesis->finishLearning();
}
catch (bad_cast& e) {
cerr << "The weak learner must be a StochasticLearner!!!" << endl;
exit(-1);
}
}
else
{
filter( pTrainingData, currentNumberOfUsedData );
if ( pTrainingData->getNumExamples() < 2 )
{
filter( pTrainingData, currentNumberOfUsedData, false );
}
if (_verbose > 1)
{
cout << "--> Size of training data = " << pTrainingData->getNumExamples() << endl;
}
energy = pWeakHypothesis->run();
pConstantWeakHypothesis->run();
}
//estimate edge
filter( pTrainingData, currentNumberOfUsedData, false );
edge = pWeakHypothesis->getEdge(true) / 2.0;
constantEdge = pConstantWeakHypothesis->getEdge() / 2.0;
if ( constantEdge > edge )
{
delete pWeakHypothesis;
pWeakHypothesis = pConstantWeakHypothesis;
edge = constantEdge;
} else {
delete pConstantWeakHypothesis;
}
// calculate alpha
AlphaReal alpha = 0.0;
alpha = 0.5 * log( ( 1 + edge ) / ( 1 - edge ) );
pWeakHypothesis->setAlpha( alpha );
_sumAlpha += alpha;
if (_verbose > 1)
cout << "Weak learner: " << pWeakHypothesis->getName()<< endl;
// Output the step-by-step information
pTrainingData->clearIndexSet();
printOutputInfo(pOutInfo, t, pTrainingData, pTestData, pWeakHypothesis);
// Updates the weights and returns the edge
//AlphaReal gamma = updateWeights(pTrainingData, pWeakHypothesis);
if (_verbose > 1)
{
cout << setprecision(5)
<< "--> Alpha = " << pWeakHypothesis->getAlpha() << endl
<< "--> Edge = " << edge << endl
<< "--> Energy = " << energy << endl
// << "--> ConstantEnergy = " << constantEnergy << endl
// << "--> difference = " << (energy - constantEnergy) << endl
;
}
// update the margins
//saveMargins();
updateMargins( pTrainingData, pWeakHypothesis );
// append the current weak learner to strong hypothesis file,
// that is, serialize it.
ss.appendHypothesis(t, pWeakHypothesis);
// Add it to the internal list of weak hypotheses
_foundHypotheses.push_back(pWeakHypothesis);
// check if the time limit has been reached
if (_maxTime > 0)
{
time( ¤tTime );
float diff = difftime(currentTime, startTime); // difftime is in seconds
diff /= 60; // = minutes
if (diff > _maxTime)
{
if (_verbose > 0)
cout << "Time limit of " << _maxTime
<< " minutes has been reached!" << endl;
break;
}
} // check for maxtime
delete pWeakHypothesis;
} // loop on iterations
/////////////////////////////////////////////////////////
// write the footer of the strong hypothesis file
ss.writeFooter();
// Free the two input data objects
if (pTrainingData)
delete pTrainingData;
if (pTestData)
delete pTestData;
if (pOutInfo)
delete pOutInfo;
if (_verbose > 0)
cout << "Learning completed." << endl;
}
void MDDAGClassifier::saveLikelihoods(const string& dataFileName, const string& shypFileName,
const string& outFileName, int numIterations)
{
InputData* pData = loadInputData(dataFileName, shypFileName);
if (_verbose > 0)
cout << "Loading strong hypothesis..." << flush;
// The class that loads the weak hypotheses
UnSerialization us;
// Where to put the weak hypotheses
vector<BaseLearner*> weakHypotheses;
// loads them
us.loadHypotheses(shypFileName, weakHypotheses, pData);
// where the results go
vector< ExampleResults* > results;
if (_verbose > 0)
cout << "Classifying..." << flush;
const int numClasses = pData->getNumClasses();
const int numExamples = pData->getNumExamples();
ofstream outFile(outFileName.c_str());
string exampleName;
if (_verbose > 0)
cout << "Output likelihoods..." << flush;
// get the results
/////////////////////////////////////////////////////////////////////
// computeResults( pData, weakHypotheses, results, numIterations );
assert( !weakHypotheses.empty() );
// Initialize the output info
OutputInfo* pOutInfo = NULL;
if ( !_outputInfoFile.empty() )
pOutInfo = new OutputInfo(_outputInfoFile, "err");
// Creating the results structures. See file Structures.h for the
// PointResults structure
results.clear();
results.reserve(numExamples);
for (int i = 0; i < numExamples; ++i)
results.push_back( new ExampleResults(i, numClasses) );
// sum votes for classes
vector< AlphaReal > votesForExamples( numClasses );
vector< AlphaReal > expVotesForExamples( numClasses );
// iterator over all the weak hypotheses
vector<BaseLearner*>::const_iterator whyIt;
int t;
pOutInfo->initialize( pData );
// for every feature: 1..T
for (whyIt = weakHypotheses.begin(), t = 0;
whyIt != weakHypotheses.end() && t < numIterations; ++whyIt, ++t)
{
BaseLearner* currWeakHyp = *whyIt;
AlphaReal alpha = currWeakHyp->getAlpha();
// for every point
for (int i = 0; i < numExamples; ++i)
{
// a reference for clarity and speed
vector<AlphaReal>& currVotesVector = results[i]->getVotesVector();
// for every class
for (int l = 0; l < numClasses; ++l)
currVotesVector[l] += alpha * currWeakHyp->classify(pData, i, l);
}
// if needed output the step-by-step information
if ( pOutInfo )
{
pOutInfo->outputIteration(t);
pOutInfo->outputCustom(pData, currWeakHyp);
// Margins and edge requires an update of the weight,
// therefore I keep them out for the moment
//outInfo.outputMargins(pData, currWeakHyp);
//outInfo.outputEdge(pData, currWeakHyp);
pOutInfo->endLine();
} // for (int i = 0; i < numExamples; ++i)
// calculate likelihoods from votes
fill( votesForExamples.begin(), votesForExamples.end(), 0.0 );
AlphaReal lLambda = 0.0;
for (int i = 0; i < numExamples; ++i)
{
// a reference for clarity and speed
vector<AlphaReal>& currVotesVector = results[i]->getVotesVector();
AlphaReal sumExp = 0.0;
// for every class
for (int l = 0; l < numClasses; ++l)
{
expVotesForExamples[l] = exp( currVotesVector[l] ) ;
sumExp += expVotesForExamples[l];
}
if ( sumExp > numeric_limits<AlphaReal>::epsilon() )
{
for (int l = 0; l < numClasses; ++l)
{
expVotesForExamples[l] /= sumExp;
}
}
Example ex = pData->getExample( results[i]->getIdx() );
vector<Label> labs = ex.getLabels();
AlphaReal m = numeric_limits<AlphaReal>::infinity();
for (int l = 0; l < numClasses; ++l)
{
if ( labs[l].y > 0 )
{
if ( expVotesForExamples[l] > numeric_limits<AlphaReal>::epsilon() )
{
AlphaReal logVal = log( expVotesForExamples[l] );
if ( logVal != m ) {
lLambda += ( ( 1.0/(AlphaReal)numExamples ) * logVal );
}
}
}
}
}
outFile << t << "\t" << lLambda ;
outFile << '\n';
outFile.flush();
}
if (pOutInfo)
delete pOutInfo;
// computeResults( pData, weakHypotheses, results, numIterations );
///////////////////////////////////////////////////////////////////////////////////
/*
for (int i = 0; i < numExamples; ++i)
{
// output the name if it exists, otherwise the number
// of the example
exampleName = pData->getExampleName(i);
if ( !exampleName.empty() )
outFile << exampleName << ',';
// output the posteriors
outFile << results[i]->getVotesVector()[0];
for (int l = 1; l < numClasses; ++l)
outFile << ',' << results[i]->getVotesVector()[l];
outFile << '\n';
}
*/
if (_verbose > 0)
cout << "Done!" << endl;
if (_verbose > 1)
{
cout << "\nClass order (You can change it in the header of the data file):" << endl;
for (int l = 0; l < numClasses; ++l)
cout << "- " << pData->getClassMap().getNameFromIdx(l) << endl;
}
// delete the input data file
if (pData)
delete pData;
vector<ExampleResults*>::iterator it;
for (it = results.begin(); it != results.end(); ++it)
delete (*it);
}
void AdaBoostMHLearner::run(const nor_utils::Args& args)
{
// load the arguments
this->getArgs(args);
// get the registered weak learner (type from name)
BaseLearner* pWeakHypothesisSource =
BaseLearner::RegisteredLearners().getLearner(_baseLearnerName);
// initialize learning options; normally it's done in the strong loop
// also, here we do it for Product learners, so input data can be created
pWeakHypothesisSource->initLearningOptions(args);
BaseLearner* pConstantWeakHypothesisSource =
BaseLearner::RegisteredLearners().getLearner("ConstantLearner");
// get the training input data, and load it
InputData* pTrainingData = pWeakHypothesisSource->createInputData();
pTrainingData->initOptions(args);
pTrainingData->load(_trainFileName, IT_TRAIN, _verbose);
// get the testing input data, and load it
InputData* pTestData = NULL;
if ( !_testFileName.empty() )
{
pTestData = pWeakHypothesisSource->createInputData();
pTestData->initOptions(args);
pTestData->load(_testFileName, IT_TEST, _verbose);
}
// The output information object
OutputInfo* pOutInfo = NULL;
if ( !_outputInfoFile.empty() )
{
// Baseline: constant classifier - goes into 0th iteration
BaseLearner* pConstantWeakHypothesis = pConstantWeakHypothesisSource->create() ;
pConstantWeakHypothesis->initLearningOptions(args);
pConstantWeakHypothesis->setTrainingData(pTrainingData);
AlphaReal constantEnergy = pConstantWeakHypothesis->run();
//pOutInfo = new OutputInfo(_outputInfoFile);
pOutInfo = new OutputInfo(args);
pOutInfo->initialize(pTrainingData);
if (pTestData)
pOutInfo->initialize(pTestData);
pOutInfo->outputHeader(pTrainingData->getClassMap());
pOutInfo->outputIteration(-1);
pOutInfo->outputCustom(pTrainingData, pConstantWeakHypothesis);
if (pTestData != NULL)
{
pOutInfo->separator();
pOutInfo->outputCustom(pTestData, pConstantWeakHypothesis);
}
pOutInfo->outputCurrentTime();
pOutInfo->endLine();
pOutInfo->initialize(pTrainingData);
if (pTestData)
pOutInfo->initialize(pTestData);
}
//cout << "Before serialization" << endl;
// reload the previously found weak learners if -resume is set.
// otherwise just return 0
int startingIteration = resumeWeakLearners(pTrainingData);
Serialization ss(_shypFileName, _isShypCompressed );
ss.writeHeader(_baseLearnerName); // this must go after resumeProcess has been called
// perform the resuming if necessary. If not it will just return
resumeProcess(ss, pTrainingData, pTestData, pOutInfo);
if (_verbose == 1)
cout << "Learning in progress..." << endl;
//I put here the starting time, but it may take very long time to load the saved model
time_t startTime, currentTime;
time(&startTime);
///////////////////////////////////////////////////////////////////////
// Starting the AdaBoost main loop
///////////////////////////////////////////////////////////////////////
for (int t = startingIteration; t < _numIterations; ++t)
{
if (_verbose > 1)
cout << "------- WORKING ON ITERATION " << (t+1) << " -------" << endl;
BaseLearner* pWeakHypothesis = pWeakHypothesisSource->create();
pWeakHypothesis->initLearningOptions(args);
//pTrainingData->clearIndexSet();
pWeakHypothesis->setTrainingData(pTrainingData);
AlphaReal energy = pWeakHypothesis->run();
//float gamma = pWeakHypothesis->getEdge();
//cout << gamma << endl;
if ( (_withConstantLearner) || ( energy != energy ) ) // check constant learner if user wants it (if energi is nan, then we chose constant learner
{
BaseLearner* pConstantWeakHypothesis = pConstantWeakHypothesisSource->create() ;
pConstantWeakHypothesis->initLearningOptions(args);
pConstantWeakHypothesis->setTrainingData(pTrainingData);
AlphaReal constantEnergy = pConstantWeakHypothesis->run();
if ( (constantEnergy <= energy) || ( energy != energy ) ) {
delete pWeakHypothesis;
pWeakHypothesis = pConstantWeakHypothesis;
}
}
if (_verbose > 1)
cout << "Weak learner: " << pWeakHypothesis->getName()<< endl;
// Output the step-by-step information
printOutputInfo(pOutInfo, t, pTrainingData, pTestData, pWeakHypothesis);
// Updates the weights and returns the edge
AlphaReal gamma = updateWeights(pTrainingData, pWeakHypothesis);
if (_verbose > 1)
{
cout << setprecision(5)
<< "--> Alpha = " << pWeakHypothesis->getAlpha() << endl
<< "--> Edge = " << gamma << endl
<< "--> Energy = " << energy << endl
// << "--> ConstantEnergy = " << constantEnergy << endl
// << "--> difference = " << (energy - constantEnergy) << endl
;
}
// If gamma <= theta the algorithm must stop.
// If theta == 0 and gamma is 0, it means that the weak learner is no better than chance
// and no further training is possible.
if (gamma <= _theta)
{
if (_verbose > 0)
{
cout << "Can't train any further: edge = " << gamma
<< " (with and edge offset (theta)=" << _theta << ")" << endl;
}
// delete pWeakHypothesis;
// break;
}
// append the current weak learner to strong hypothesis file,
// that is, serialize it.
ss.appendHypothesis(t, pWeakHypothesis);
// Add it to the internal list of weak hypotheses
_foundHypotheses.push_back(pWeakHypothesis);
// check if the time limit has been reached
if (_maxTime > 0)
{
time( ¤tTime );
float diff = difftime(currentTime, startTime); // difftime is in seconds
diff /= 60; // = minutes
if (diff > _maxTime)
{
if (_verbose > 0)
cout << "Time limit of " << _maxTime
<< " minutes has been reached!" << endl;
break;
}
} // check for maxtime
delete pWeakHypothesis;
} // loop on iterations
/////////////////////////////////////////////////////////
// write the footer of the strong hypothesis file
ss.writeFooter();
// write the weights of the instances if the name of weights file isn't empty
printOutWeights( pTrainingData );
// Free the two input data objects
if (pTrainingData)
delete pTrainingData;
if (pTestData)
delete pTestData;
if (pOutInfo)
delete pOutInfo;
if (_verbose > 0)
cout << "Learning completed." << endl;
}