void BanditSingleStumpLearner::declareArguments(nor_utils::Args& args)
{
FeaturewiseLearner::declareArguments(args);
args.declareArgument("updaterule",
"The update weights in the UCT can be the 1-sqrt( 1- edge^2 ) [edge]\n"
" or the alpha [alphas]\n"
" Default is the first one\n",
1, "<type>");
args.declareArgument("rsample",
"Number of features to be considered\n"
" Default is one\n",
1, "<K>");
args.declareArgument("banditalgo",
"The bandit algorithm (UCBK, UCBKRandomized, EXP3 )\n"
"Default is UCBK\n",
1, "<algoname>");
args.declareArgument("percent",
"The percent of database will be used for estimating the payoffs(EXP3G)\n"
" Default is 10%\n",
1, "<p>");
}
void StochasticLearner::declareArguments(nor_utils::Args& args)
{
BaseLearner::declareArguments(args);
args.declareArgument("graditer",
"Declares the number of randomly drawn training size for SGD"
"whereas it declares the number of iteration for the Batch Gradiend Descend"
" size <num> of training set. "
"Example: --graditer 50 -> Uses only 50 randomly chosen training instance",
1, "<num>");
args.declareArgument("gradmethod",
"Declares the gradient method: "
" (sgd) Stochastic Gradient Descent, (bgd) Batch Gradient Descent"
"Example: --gradmethod sgd -> Uses stochastic gradient method",
1, "<method>");
args.declareArgument("tfunc",
"Target function: "
"exploss: Exponential Loss, edge: max. edge"
"Example: --tfunc exploss -> Uses exponantial loss for minimizing",
1, "<function>");
args.declareArgument("initgamma",
"The initial learning rate in gradient descent"
"Default values is 10.0",
1, "<gamma>");
args.declareArgument("gammdivperiod",
"The periodicity of decreasing the learning rate \\gamma"
"Default values is 1",
1, "<period>");
}
MDDAGClassifier::MDDAGClassifier(const nor_utils::Args &args, int verbose)
: _verbose(verbose), _args(args)
{
// The file with the step-by-step information
if ( args.hasArgument("outputinfo") )
args.getValue("outputinfo", 0, _outputInfoFile);
}
void TreeLearner::initLearningOptions(const nor_utils::Args& args)
{
BaseLearner::initLearningOptions(args);
string baseLearnerName;
args.getValue("baselearnertype", 0, baseLearnerName);
args.getValue("baselearnertype", 1, _numBaseLearners);
// get the registered weak learner (type from name)
BaseLearner* pWeakHypothesisSource = BaseLearner::RegisteredLearners().getLearner(baseLearnerName);
//check whether the weak learner is a ScalarLeaerner
try {
_pScalaWeakHypothesisSource = dynamic_cast<ScalarLearner*>(pWeakHypothesisSource);
}
catch (bad_cast& e) {
cerr << "The weak hypothesis must be a ScalarLearner!!!" << endl;
exit(-1);
}
_pScalaWeakHypothesisSource->initLearningOptions(args);
/*
for( int ib = 0; ib < _numBaseLearners; ++ib ) {
vector< int > tmpVector( 2, -1 );
_idxPairs.push_back( tmpVector );
}
*/
}
void EnumLearnerSA::initLearningOptions(const nor_utils::Args& args)
{
BaseLearner::initLearningOptions(args);
if ( args.hasArgument( "uoffset" ) )
args.getValue("uoffset", 0, _uOffset);
}
void MultiMDDAGLearner::getArgs(const nor_utils::Args& args)
{
MDDAGLearner::getArgs(args);
// Set the value of theta
if ( args.hasArgument("updateperc") )
args.getValue("updateperc", 0, _randomNPercent);
}
void BaseLearner::initLearningOptions(const nor_utils::Args& args)
{
if ( args.hasArgument("verbose") )
args.getValue("verbose", 0, _verbose);
// Set the value of theta
if ( args.hasArgument("edgeoffset") )
args.getValue("edgeoffset", 0, _theta);
}
void ParasiteLearner::initLearningOptions(const nor_utils::Args& args)
{
BaseLearner::initLearningOptions(args);
args.getValue("pool", 0, _nameBaseLearnerFile);
args.getValue("pool", 1, _numBaseLearners);
if ( args.hasArgument("closed") )
_closed = 1;
}
void FilterBoostLearner::getArgs(const nor_utils::Args& args)
{
AdaBoostMHLearner::getArgs( args );
// Set the value of the sample size
if ( args.hasArgument("Cn") )
{
args.getValue("C", 0, _Cn);
if (_verbose > 1)
cout << "--> Resampling size: " << _Cn << endl;
}
}
int MultiMDDAGLearner::resumeProcess(const nor_utils::Args& args, InputData* pTestData)
{
int numPolicies = 0;
AlphaReal policyAlpha = 0.0;
if ( args.hasArgument("policyalpha") )
args.getValue("policyalpha", 0, policyAlpha);
_policy = new AdaBoostArrayOfPolicyArray(args, _actionNumber);
return numPolicies;
}
void SoftCascadeLearner::classify(const nor_utils::Args& args)
{
SoftCascadeClassifier classifier(args, _verbose);
string testFileName = args.getValue<string>("test", 0);
string shypFileName = args.getValue<string>("test", 1);
int numIterations = args.getValue<int>("test", 2);
string outResFileName = "";
if ( args.getNumValues("test") > 3 )
args.getValue("test", 3, outResFileName);
classifier.run(testFileName, shypFileName, numIterations, outResFileName);
}
void ParasiteLearner::declareArguments(nor_utils::Args& args)
{
BaseLearner::declareArguments(args);
args.declareArgument("pool",
"The name of the shyp file containing the pool of\n"
" weak learners, followed by the number of desired\n"
" weak learners. If -1 or more than the number of \n"
" weak learners, we use all of them",
2, "<fileName> <nBaseLearners>");
args.declareArgument("closed", "Include negatives of weak learners (default = false).");
}
void FilterBoostLearner::classify(const nor_utils::Args& args)
{
FilterBoostClassifier classifier(args, _verbose);
// -test <dataFile> <shypFile>
string testFileName = args.getValue<string>("test", 0);
string shypFileName = args.getValue<string>("test", 1);
int numIterations = args.getValue<int>("test", 2);
string outResFileName;
if ( args.getNumValues("test") > 3 )
args.getValue("test", 3, outResFileName);
classifier.run(testFileName, shypFileName, numIterations, outResFileName);
}
VJCascadeClassifier::VJCascadeClassifier(const nor_utils::Args &args, int verbose)
: _verbose(verbose), _args(args), _positiveLabelIndex(-1)
{
// The file with the step-by-step information
if ( args.hasArgument("outputinfo") )
args.getValue("outputinfo", 0, _outputInfoFile);
if ( args.hasArgument("positivelabel") )
{
args.getValue("positivelabel", 0, _positiveLabelName);
} else {
cout << "The name of positive label has to be given!!!" << endl;
exit(-1);
}
}
//----------------------------------------------------------------
//----------------------------------------------------------------
void Exp3::initLearningOptions(const nor_utils::Args& args)
{
if ( args.hasArgument( "gamma" ) ){
_gamma = args.getValue<double>("gamma", 0 );
}
}
/**
* Show the help. Called when -h argument is provided.
* \date 11/11/2005
*/
void showHelp(nor_utils::Args& args, const vector<string>& learnersList)
{
cout << "MultiBoost (v" << CURRENT_VERSION << "). An obvious name for a multi-class AdaBoost learner." << endl;
cout << "------------------------ HELP SECTION --------------------------" << endl;
args.printGroup("Parameters");
cout << endl;
cout << "For specific help options type:" << endl;
cout << " --h general: General options" << endl;
cout << " --h io: I/O options" << endl;
cout << " --h algo: Basic algorithm options" << endl;
cout << " --h bandits: Bandit algorithm options" << endl;
cout << " --h vjcascade: Viola-Jones Cascade options" << endl;
cout << " --h softcascade: Soft Cascade options" << endl;
cout << endl;
cout << "For weak learners specific options type:" << endl;
vector<string>::const_iterator it;
for (it = learnersList.begin(); it != learnersList.end(); ++it)
cout << " --h " << *it << endl;
exit(0);
}
void StochasticLearner::initLearningOptions(const nor_utils::Args& args)
{
BaseLearner::initLearningOptions(args);
if (args.hasArgument("initgamma"))
args.getValue("initgamma", 0, _initialGammat);
if (args.hasArgument("gammdivperiod"))
args.getValue("gammdivperiod", 0, _gammdivperiod);
if (args.hasArgument("graditer"))
args.getValue("graditer", 0, _maxIter);
if (args.hasArgument("gradmethod"))
{
string gradMethod;
args.getValue("gradmethod", 0, gradMethod);
if ( gradMethod.compare( "sgd" ) == 0 )
_gMethod = OPT_SGD;
else if ( gradMethod.compare( "bgd" ) == 0 )
_gMethod = OPT_BGD;
else {
cerr << "SigmoidSingleStumpLearner::Unknown update gradient method" << endl;
exit( -1 );
}
}
if (args.hasArgument("tfunc"))
{
string targetFunction;
args.getValue("tfunc", 0, targetFunction);
if ( targetFunction.compare( "exploss" ) == 0 )
_tFunction = TF_EXPLOSS;
else if ( targetFunction.compare( "edge" ) == 0 )
_tFunction = TF_EDGE;
else {
cerr << "SigmoidSingleStumpLearner::Unknown target function" << endl;
exit( -1 );
}
}
}
void EnumLearnerSA::declareArguments(nor_utils::Args& args)
{
BaseLearner::declareArguments(args);
args.declareArgument("uoffset",
"The offset of u\n",
1, "<offset>");
}
void FeaturewiseLearner::initLearningOptions(const nor_utils::Args& args)
{
AbstainableLearner::initLearningOptions(args);
_maxNumOfDimensions = numeric_limits<int>::max();
// If the sampling is required
if ( args.hasArgument("rsample") )
_maxNumOfDimensions = args.getValue<int>("rsample", 0);
}
void TreeLearnerUCT::declareArguments(nor_utils::Args& args)
{
BaseLearner::declareArguments(args);
args.declareArgument("baselearnertype",
"The name of the learner that serves as a basis for the product\n"
" and the number of base learners to be multiplied\n"
" Don't forget to add its parameters\n",
2, "<baseLearnerType> <numBaseLearners>");
args.declareArgument("updaterule",
"The update weights in the UCT can be the 1-sqrt( 1- edge^2 ) [edge]\n"
" or the alpha [alphas]\n"
" or edgesquare [edgesquare]\n"
" Default is the first one\n",
1, "<type>");
}
void ProductLearner::initLearningOptions(const nor_utils::Args& args)
{
BaseLearner::initLearningOptions(args);
string baseLearnerName;
args.getValue("baselearnertype", 0, baseLearnerName);
args.getValue("baselearnertype", 1, _numBaseLearners);
// get the registered weak learner (type from name)
BaseLearner* pWeakHypothesisSource =
BaseLearner::RegisteredLearners().getLearner(baseLearnerName);
pWeakHypothesisSource->initLearningOptions(args);
for( int ib = 0; ib < _numBaseLearners; ++ib ) {
_baseLearners.push_back(pWeakHypothesisSource->create());
_baseLearners[ib]->initLearningOptions(args);
}
}
void BanditTreeLearner::declareArguments(nor_utils::Args& args)
{
BanditLearner::declareArguments(args);
args.declareArgument("baselearnertype",
"The name of the learner that serves as a basis for the product\n"
" and the number of base learners to be multiplied\n"
" Don't forget to add its parameters\n",
2, "<baseLearnerType> <numBaseLearners>");
}
void BaseLearner::declareBaseArguments(nor_utils::Args& args)
{
args.declareArgument("shypname",
"The name of output strong hypothesis (default: "
+ string(SHYP_NAME) + "." + string(SHYP_EXTENSION) + ").",
1, "<filename>");
args.declareArgument("shypcomp",
"The shyp file will be compressed",
1, "<flag 0-1>");
args.setGroup("Basic Algorithm Options");
args.declareArgument("resume",
"Resumes a training process using the strong hypothesis file.",
1, "<shypFile>");
args.declareArgument("edgeoffset",
"Defines the value of the edge offset (theta) (default: no edge offset).",
1, "<val>");
}
void AbstainableLearner::declareArguments(nor_utils::Args& args)
{
BaseLearner::declareArguments(args);
args.declareArgument("abstention",
"Activate the abstention. Available types are:\n"
" greedy: sorting and checking in O(k^2)\n"
" full: the O(2^k) full search\n"
" real: use the AdaBoost.MH with real valued predictions\n"
" classwise: abstain if classwise edge <= theta",
1, "<type>");
}
void FeaturewiseLearner::declareArguments(nor_utils::Args& args)
{
AbstainableLearner::declareArguments(args);
args.declareArgument("rsample",
"Instead of searching for a featurewise in all the possible dimensions (features), select a set of "
" size <num> of random dimensions. "
"Example: -rsample 50 -> Search over only 50 dimensions"
"(Turned off for Haar: use -csample instead)",
1, "<num>");
}
void FilterBoostLearner::doConfusionMatrix(const nor_utils::Args& args)
{
FilterBoostClassifier classifier(args, _verbose);
// -cmatrix <dataFile> <shypFile>
if ( args.hasArgument("cmatrix") )
{
string testFileName = args.getValue<string>("cmatrix", 0);
string shypFileName = args.getValue<string>("cmatrix", 1);
classifier.printConfusionMatrix(testFileName, shypFileName);
}
// -cmatrixfile <dataFile> <shypFile> <outFile>
else if ( args.hasArgument("cmatrixfile") )
{
string testFileName = args.getValue<string>("cmatrix", 0);
string shypFileName = args.getValue<string>("cmatrix", 1);
string outResFileName = args.getValue<string>("cmatrix", 2);
classifier.saveConfusionMatrix(testFileName, shypFileName, outResFileName);
}
}
void TreeLearnerUCT::initLearningOptions(const nor_utils::Args& args)
{
BaseLearner::initLearningOptions(args);
string baseLearnerName;
args.getValue("baselearnertype", 0, baseLearnerName);
args.getValue("baselearnertype", 1, _numBaseLearners);
// get the registered weak learner (type from name)
BaseLearner* pWeakHypothesisSource =
BaseLearner::RegisteredLearners().getLearner(baseLearnerName);
for( int ib = 0; ib < _numBaseLearners; ++ib ) {
_baseLearners.push_back(pWeakHypothesisSource->create());
_baseLearners[ib]->initLearningOptions(args);
vector< int > tmpVector( 2, -1 );
_idxPairs.push_back( tmpVector );
}
string updateRule = "";
if ( args.hasArgument( "updaterule" ) )
args.getValue("updaterule", 0, updateRule );
if ( updateRule.compare( "edge" ) == 0 )
_updateRule = EDGE_SQUARE;
else if ( updateRule.compare( "alphas" ) == 0 )
_updateRule = ALPHAS;
else if ( updateRule.compare( "edgesquare" ) == 0 )
_updateRule = ESQUARE;
else {
cerr << "Unknown update rule in ProductLearnerUCT (set to default [edge]" << endl;
_updateRule = EDGE_SQUARE;
}
}
void AdaBoostMHLearner::doPosteriors(const nor_utils::Args& args)
{
AdaBoostMHClassifier classifier(args, _verbose);
int numofargs = args.getNumValues( "posteriors" );
// -posteriors <dataFile> <shypFile> <outFile> <numIters>
string testFileName = args.getValue<string>("posteriors", 0);
string shypFileName = args.getValue<string>("posteriors", 1);
string outFileName = args.getValue<string>("posteriors", 2);
int numIterations = args.getValue<int>("posteriors", 3);
int period = 0;
if ( numofargs == 5 )
period = args.getValue<int>("posteriors", 4);
classifier.savePosteriors(testFileName, shypFileName, outFileName, numIterations, period);
}
void AbstainableLearner::initLearningOptions(const nor_utils::Args& args)
{
BaseLearner::initLearningOptions(args);
// set abstention
if ( args.hasArgument("abstention") )
{
string abstType = args.getValue<string>("abstention", 0);
if (abstType == "greedy")
_abstention = ABST_GREEDY;
else if (abstType == "full")
_abstention = ABST_FULL;
else if (abstType == "real")
_abstention = ABST_REAL;
else if (abstType == "classwise")
_abstention = ABST_CLASSWISE;
else
{
cerr << "ERROR: Invalid type of abstention <" << abstType << ">!!" << endl;
exit(1);
}
}
}
/**
* Show the help for the options.
* \param args The arguments structure.
* \date 28/11/2005
*/
void showOptionalHelp(nor_utils::Args& args)
{
string helpType = args.getValue<string>("h", 0);
cout << "MultiBoost (v" << CURRENT_VERSION << "). An obvious name for a multi-class AdaBoost learner." << endl;
cout << "---------------------------------------------------------------------------" << endl;
if (helpType == "general")
args.printGroup("General Options");
else if (helpType == "io")
args.printGroup("I/O Options");
else if (helpType == "algo")
args.printGroup("Basic Algorithm Options");
else if (helpType == "bandits")
args.printGroup("Bandit Algorithm Options");
else if (helpType == "vjcascade")
args.printGroup("Viola-Jones Cascade Algorithm Options");
else if (helpType == "softcascade")
args.printGroup("SoftCascade Algorithm Options");
else if ( BaseLearner::RegisteredLearners().hasLearner(helpType) )
args.printGroup(helpType + " Options");
else
cerr << "ERROR: Unknown help section <" << helpType << ">" << endl;
}