CMSSW Reference Manual

void help ( const char * prog )

Definition at line 38 of file SprBoosterApp.cc.

References GenMuonPlsPt100GeV_cfg::cout, and lat::endl().

00039 {
00040   cout << "Usage:  " << prog 
00041        << " training_data_file"
00042        << " file_of_classifier_parameters(see booster.config for syntax)" 
00043        << endl;
00044   cout << "\t Options: " << endl;
00045   cout << "\t-h --- help                                        " << endl;
00046   cout << "\t-M AdaBoost mode                                   " << endl;
00047   cout << "\t\t 1 = Discrete AdaBoost (default)                 " << endl;
00048   cout << "\t\t 2 = Real AdaBoost                               " << endl;
00049   cout << "\t\t 3 = Epsilon AdaBoost                            " << endl;
00050   cout << "\t-E epsilon for Epsilon and Real AdaBoosts (def=0.01)" << endl;
00051   cout << "\t-a input ascii file mode (see SprSimpleReader.hh)  " << endl;
00052   cout << "\t-A save output data in ascii instead of Root       " << endl;
00053   cout << "\t-n number of AdaBoost training cycles              " << endl;
00054   cout << "\t-y list of input classes (see SprAbsFilter.hh)     " << endl;
00055   cout << "\t-Q apply variable transformation saved in file     " << endl;
00056   cout << "\t-g per-event loss for (cross-)validation           " << endl;
00057   cout << "\t\t 1 - quadratic loss (y-f(x))^2                   " << endl;
00058   cout << "\t\t 2 - exponential loss exp(-y*f(x))               " << endl;
00059   cout << "\t-b bootstrap training sample                       " << endl;
00060   cout << "\t-m replace data values below this cutoff with medians" << endl;
00061   cout << "\t-e skip initial event reweighting when resuming    " << endl;
00062   cout << "\t-u store data with modified weights to file        " << endl;
00063   cout << "\t-v verbose level (0=silent default,1,2)            " << endl;
00064   cout << "\t-f store trained AdaBoost to file                  " << endl;
00065   cout << "\t-r resume training for AdaBoost stored in file     " << endl;
00066   cout << "\t-K keep this fraction in training set and          " << endl;
00067   cout << "\t\t put the rest into validation set                " << endl;
00068   cout << "\t-D randomize training set split-up                 " << endl;
00069   cout << "\t-t read validation/test data from a file           " << endl;
00070   cout << "\t\t (must be in same format as input data!!!        " << endl;
00071   cout << "\t-d frequency of print-outs for validation data     " << endl;
00072   cout << "\t-w scale all signal weights by this factor         " << endl;
00073   cout << "\t-V include only these input variables              " << endl;
00074   cout << "\t-z exclude input variables from the list           " << endl;
00075   cout << "\t\t Variables must be listed in quotes and separated by commas." 
00076        << endl;
00077 }

int main	(	int	argc,
		char **	argv
	)

Definition at line 90 of file SprBoosterApp.cc.

References c, TestMuL1L2Filter_cff::cerr, GenMuonPlsPt100GeV_cfg::cout, lat::endl(), HLT_VtxMuL3::Epsilon, geometryDiff::epsilon, file, filter, help(), i, prepareExit(), split, t, vars, and weights.

00091 {
00092   // check command line
00093   if( argc < 3 ) {
00094     help(argv[0]);
00095     return 1;
00096   }
00097 
00098   // init
00099   int readMode = 0;
00100   SprRWFactory::DataType writeMode = SprRWFactory::Root;
00101   unsigned cycles = 0;
00102   int verbose = 0;
00103   string outFile;
00104   string resumeFile;
00105   string valFile;
00106   unsigned valPrint = 0;
00107   bool scaleWeights = false;
00108   double sW = 1.;
00109   bool useStandardAB = false;
00110   int iAdaBoostMode = 1;
00111   double epsilon = 0.01;
00112   bool skipInitialEventReweighting = false;
00113   string weightedDataOut;
00114   bool setLowCutoff = false;
00115   double lowCutoff = 0;
00116   string includeList, excludeList;
00117   int iLoss = 0;
00118   bool bagInput = false;
00119   string inputClassesString;
00120   bool split = false;
00121   double splitFactor = 0;
00122   bool splitRandomize = false;
00123   string transformerFile;
00124 
00125   // decode command line
00126   int c;
00127   extern char* optarg;
00128   //  extern int optind;
00129   while((c = getopt(argc,argv,"hM:E:a:An:y:Q:g:bm:eu:v:f:r:K:Dt:d:w:V:z:")) != EOF ) {
00130     switch( c )
00131       {
00132       case 'h' :
00133         help(argv[0]);
00134         return 1;
00135       case 'M' :
00136         iAdaBoostMode = (optarg==0 ? 1 : atoi(optarg));
00137         break;
00138       case 'E' :
00139         epsilon = (optarg==0 ? 0.01 : atof(optarg));
00140         break;
00141       case 'a' :
00142         readMode = (optarg==0 ? 0 : atoi(optarg));
00143         break;
00144       case 'A' :
00145         writeMode = SprRWFactory::Ascii;
00146         break;
00147       case 'n' :
00148         cycles = (optarg==0 ? 1 : atoi(optarg));
00149         break;
00150       case 'y' :
00151         inputClassesString = optarg;
00152         break;
00153       case 'g' :
00154         iLoss = (optarg==0 ? 0 : atoi(optarg));
00155         break;
00156       case 'b' :
00157         bagInput = true;
00158         break;
00159       case 'm' :
00160         if( optarg != 0 ) {
00161           setLowCutoff = true;
00162           lowCutoff = atof(optarg);
00163         }
00164         break;
00165       case 'e' :
00166         skipInitialEventReweighting = true;
00167         break;
00168       case 'u' :
00169         weightedDataOut = optarg;
00170         break;
00171       case 'v' :
00172         verbose = (optarg==0 ? 0 : atoi(optarg));
00173         break;
00174       case 'f' :
00175         outFile = optarg;
00176         break;
00177       case 'r' :
00178         resumeFile = optarg;
00179         break;
00180       case 'K' :
00181         split = true;
00182         splitFactor = (optarg==0 ? 0 : atof(optarg));
00183         break;
00184       case 'D' :
00185         splitRandomize = true;
00186         break;
00187       case 't' :
00188         valFile = optarg;
00189         break;
00190       case 'd' :
00191         valPrint = (optarg==0 ? 0 : atoi(optarg));
00192         break;
00193       case 'w' :
00194         if( optarg != 0 ) {
00195           scaleWeights = true;
00196           sW = atof(optarg);
00197         }
00198         break;
00199       case 'V' :
00200         includeList = optarg;
00201         break;
00202       case 'z' :
00203         excludeList = optarg;
00204         break;
00205       }
00206   }
00207 
00208   // Must have 2 arguments after all options.
00209   string trFile = argv[argc-2];
00210   string configFile = argv[argc-1]; 
00211   if( trFile.empty() ) {
00212     cerr << "No training file is specified." << endl;
00213     return 1;
00214   }
00215   if( configFile.empty() ) {
00216     cerr << "No classifier configuration file specified." << endl;
00217     return 1;
00218   }
00219 
00220   // make reader
00221   SprRWFactory::DataType inputType 
00222     = ( readMode==0 ? SprRWFactory::Root : SprRWFactory::Ascii );
00223   auto_ptr<SprAbsReader> reader(SprRWFactory::makeReader(inputType,readMode));
00224 
00225   // include variables
00226   set<string> includeSet;
00227   if( !includeList.empty() ) {
00228     vector<vector<string> > includeVars;
00229     SprStringParser::parseToStrings(includeList.c_str(),includeVars);
00230     assert( !includeVars.empty() );
00231     for( int i=0;i<includeVars[0].size();i++ ) 
00232       includeSet.insert(includeVars[0][i]);
00233     if( !reader->chooseVars(includeSet) ) {
00234       cerr << "Unable to include variables in training set." << endl;
00235       return 2;
00236     }
00237     else {
00238       cout << "Following variables have been included in optimization: ";
00239       for( set<string>::const_iterator 
00240              i=includeSet.begin();i!=includeSet.end();i++ )
00241         cout << "\"" << *i << "\"" << " ";
00242       cout << endl;
00243     }
00244   }
00245 
00246   // exclude variables
00247   set<string> excludeSet;
00248   if( !excludeList.empty() ) {
00249     vector<vector<string> > excludeVars;
00250     SprStringParser::parseToStrings(excludeList.c_str(),excludeVars);
00251     assert( !excludeVars.empty() );
00252     for( int i=0;i<excludeVars[0].size();i++ ) 
00253       excludeSet.insert(excludeVars[0][i]);
00254     if( !reader->chooseAllBut(excludeSet) ) {
00255       cerr << "Unable to exclude variables from training set." << endl;
00256       return 2;
00257     }
00258     else {
00259       cout << "Following variables have been excluded from optimization: ";
00260       for( set<string>::const_iterator 
00261              i=excludeSet.begin();i!=excludeSet.end();i++ )
00262         cout << "\"" << *i << "\"" << " ";
00263       cout << endl;
00264     }
00265   }
00266 
00267   // read training data from file
00268   auto_ptr<SprAbsFilter> filter(reader->read(trFile.c_str()));
00269   if( filter.get() == 0 ) {
00270     cerr << "Unable to read data from file " << trFile.c_str() << endl;
00271     return 2;
00272   }
00273   vector<string> vars;
00274   filter->vars(vars);
00275   cout << "Read data from file " << trFile.c_str() << " for variables";
00276   for( int i=0;i<vars.size();i++ ) 
00277     cout << " \"" << vars[i].c_str() << "\"";
00278   cout << endl;
00279   cout << "Total number of points read: " << filter->size() << endl;
00280 
00281   // filter training data by class
00282   vector<SprClass> inputClasses;
00283   if( !filter->filterByClass(inputClassesString.c_str()) ) {
00284     cerr << "Cannot choose input classes for string " 
00285          << inputClassesString << endl;
00286     return 2;
00287   }
00288   filter->classes(inputClasses);
00289   assert( inputClasses.size() > 1 );
00290   cout << "Training data filtered by class." << endl;
00291   for( int i=0;i<inputClasses.size();i++ ) {
00292     cout << "Points in class " << inputClasses[i] << ":   " 
00293          << filter->ptsInClass(inputClasses[i]) << endl;
00294   }
00295 
00296   // scale weights
00297   if( scaleWeights ) {
00298     cout << "Signal weights are multiplied by " << sW << endl;
00299     filter->scaleWeights(inputClasses[1],sW);
00300   }
00301 
00302   // apply low cutoff
00303   if( setLowCutoff ) {
00304     if( !filter->replaceMissing(SprUtils::lowerBound(lowCutoff),1) ) {
00305       cerr << "Unable to replace missing values in training data." << endl;
00306       return 2;
00307     }
00308     else
00309       cout << "Values below " << lowCutoff << " in training data"
00310            << " have been replaced with medians." << endl;
00311   }
00312 
00313   // read validation data from file
00314   auto_ptr<SprAbsFilter> valFilter;
00315   if( split && !valFile.empty() ) {
00316     cerr << "Unable to split training data and use validation data " 
00317          << "from a separate file." << endl;
00318     return 2;
00319   }
00320   if( split && valPrint!=0 ) {
00321     cout << "Splitting training data with factor " << splitFactor << endl;
00322     if( splitRandomize )
00323       cout << "Will use randomized splitting." << endl;
00324     vector<double> weights;
00325     SprData* splitted = filter->split(splitFactor,weights,splitRandomize);
00326     if( splitted == 0 ) {
00327       cerr << "Unable to split training data." << endl;
00328       return 2;
00329     }
00330     bool ownData = true;
00331     valFilter.reset(new SprEmptyFilter(splitted,weights,ownData));
00332     cout << "Training data re-filtered:" << endl;
00333     for( int i=0;i<inputClasses.size();i++ ) {
00334       cout << "Points in class " << inputClasses[i] << ":   " 
00335            << filter->ptsInClass(inputClasses[i]) << endl;
00336     }
00337   }
00338   if( !valFile.empty() && valPrint!=0 ) {
00339     auto_ptr<SprAbsReader> 
00340       valReader(SprRWFactory::makeReader(inputType,readMode));
00341     if( !includeSet.empty() ) {
00342       if( !valReader->chooseVars(includeSet) ) {
00343         cerr << "Unable to include variables in validation set." << endl;
00344         return 2;
00345       }
00346     }
00347     if( !excludeSet.empty() ) {
00348       if( !valReader->chooseAllBut(excludeSet) ) {
00349         cerr << "Unable to exclude variables from validation set." << endl;
00350         return 2;
00351       }
00352     }
00353     valFilter.reset(valReader->read(valFile.c_str()));
00354     if( valFilter.get() == 0 ) {
00355       cerr << "Unable to read data from file " << valFile.c_str() << endl;
00356       return 2;
00357     }
00358     vector<string> valVars;
00359     valFilter->vars(valVars);
00360     cout << "Read validation data from file " << valFile.c_str() 
00361          << " for variables";
00362     for( int i=0;i<valVars.size();i++ ) 
00363       cout << " \"" << valVars[i].c_str() << "\"";
00364     cout << endl;
00365     cout << "Total number of points read: " << valFilter->size() << endl;
00366     cout << "Points in class 0: " << valFilter->ptsInClass(inputClasses[0])
00367          << " 1: " << valFilter->ptsInClass(inputClasses[1]) << endl;
00368   }
00369 
00370   // filter validation data by class
00371   if( valFilter.get() != 0 ) {
00372     if( !valFilter->filterByClass(inputClassesString.c_str()) ) {
00373       cerr << "Cannot choose input classes for string " 
00374            << inputClassesString << endl;
00375       return 2;
00376     }
00377     valFilter->classes(inputClasses);
00378     cout << "Validation data filtered by class." << endl;
00379     for( int i=0;i<inputClasses.size();i++ ) {
00380       cout << "Points in class " << inputClasses[i] << ":   " 
00381            << valFilter->ptsInClass(inputClasses[i]) << endl;
00382     }
00383   }
00384 
00385   // scale weights
00386   if( scaleWeights && valFilter.get()!=0 )
00387     valFilter->scaleWeights(inputClasses[1],sW);
00388 
00389   // apply low cutoff
00390   if( setLowCutoff && valFilter.get()!=0 ) {
00391     if( !valFilter->replaceMissing(SprUtils::lowerBound(lowCutoff),1) ) {
00392       cerr << "Unable to replace missing values in validation data." << endl;
00393       return 2;
00394     }
00395     else
00396       cout << "Values below " << lowCutoff << " in validation data"
00397            << " have been replaced with medians." << endl;
00398   }
00399 
00400   // apply transformation of variables to training and test data
00401   auto_ptr<SprAbsFilter> garbage_train, garbage_valid;
00402   if( !transformerFile.empty() ) {
00403     SprVarTransformerReader transReader;
00404     const SprAbsVarTransformer* t = transReader.read(transformerFile.c_str());
00405     if( t == 0 ) {
00406       cerr << "Unable to read VarTransformer from file "
00407            << transformerFile.c_str() << endl;
00408       return 2;
00409     }
00410     SprTransformerFilter* t_train = new SprTransformerFilter(filter.get());
00411     SprTransformerFilter* t_valid = 0;
00412     if( valFilter.get() != 0 )
00413       t_valid = new SprTransformerFilter(valFilter.get());
00414     bool replaceOriginalData = true;
00415     if( !t_train->transform(t,replaceOriginalData) ) {
00416       cerr << "Unable to apply VarTransformer to training data." << endl;
00417       return 2;
00418     }
00419     if( t_valid!=0 && !t_valid->transform(t,replaceOriginalData) ) {
00420       cerr << "Unable to apply VarTransformer to validation data." << endl;
00421       return 2;
00422     }
00423     cout << "Variable transformation from file "
00424          << transformerFile.c_str() << " has been applied to "
00425          << "training and validation data." << endl;
00426     garbage_train.reset(filter.release());
00427     garbage_valid.reset(valFilter.release());
00428     filter.reset(t_train);
00429     valFilter.reset(t_valid);
00430   }
00431 
00432   // make per-event loss
00433   auto_ptr<SprAverageLoss> loss;
00434   switch( iLoss )
00435     {
00436     case 1 :
00437       loss.reset(new SprAverageLoss(&SprLoss::quadratic,
00438                                     &SprTransformation::logit));
00439       cout << "Per-event loss set to "
00440            << "Quadratic loss (y-f(x))^2 " << endl;
00441       break;
00442     case 2 :
00443       loss.reset(new SprAverageLoss(&SprLoss::exponential));
00444       cout << "Per-event loss set to "
00445            << "Exponential loss exp(-y*f(x)) " << endl;
00446       break;
00447     default :
00448       cout << "No per-event loss is chosen. Will use the default." << endl;
00449       break;
00450     }
00451 
00452   // make AdaBoost mode
00453   SprTrainedAdaBoost::AdaBoostMode abMode = SprTrainedAdaBoost::Discrete;
00454   switch( iAdaBoostMode )
00455     {
00456     case 1 :
00457       abMode = SprTrainedAdaBoost::Discrete;
00458       cout << "Will train Discrete AdaBoost." << endl;
00459       break;
00460     case 2 :
00461       abMode = SprTrainedAdaBoost::Real;
00462       cout << "Will train Real AdaBoost." << endl;
00463       break;
00464     case 3 :
00465       abMode = SprTrainedAdaBoost::Epsilon;
00466       cout << "Will train Epsilon AdaBoost." << endl;
00467       break;
00468    default :
00469       cout << "Will train Discrete AdaBoost." << endl;
00470       break;
00471     }
00472 
00473   // open file with classifier configs
00474   ifstream file(configFile.c_str());
00475   if( !file ) {
00476     cerr << "Unable to open file " << configFile.c_str() << endl;
00477     return 3;
00478   }
00479 
00480   // prepare vectors of objects
00481   vector<SprAbsTwoClassCriterion*> criteria;
00482   vector<SprAbsClassifier*> destroyC;// classifiers to be deleted
00483   vector<SprIntegerBootstrap*> bstraps;
00484   vector<SprCCPair> useC;// classifiers and cuts to be used
00485 
00486   // read classifier params
00487   unsigned nLine = 0;
00488   bool discreteTree = (abMode!=SprTrainedAdaBoost::Real);
00489   bool mixedNodesTree = (abMode==SprTrainedAdaBoost::Real);
00490   bool fastSort = true;
00491   bool readOneEntry = false;
00492   if( !SprClassifierReader::readTrainableConfig(file,nLine,filter.get(),
00493                                                 discreteTree,mixedNodesTree,
00494                                                 fastSort,criteria,
00495                                                 bstraps,destroyC,useC,
00496                                                 readOneEntry) ) {
00497     cerr << "Unable to read weak classifier configurations from file " 
00498          << configFile.c_str() << endl;
00499     prepareExit(criteria,destroyC,bstraps);
00500     return 4;
00501   }
00502   cout << "Finished reading " << useC.size() << " classifiers from file "
00503        << configFile.c_str() << endl;
00504 
00505   // make AdaBoost
00506   SprAdaBoost ab(filter.get(),cycles,useStandardAB,abMode,bagInput);
00507   cout << "Setting epsilon to " << epsilon << endl;
00508   ab.setEpsilon(epsilon);
00509 
00510   // set validation
00511   if( valFilter.get()!=0 && !valFilter->empty() )
00512     ab.setValidation(valFilter.get(),valPrint,loss.get());
00513 
00514   // read saved Boost from file
00515   if( !resumeFile.empty() ) {
00516     if( !SprClassifierReader::readTrainable(resumeFile.c_str(),
00517                                             &ab,verbose) ) {
00518       cerr << "Failed to read saved AdaBoost from file " 
00519            << resumeFile.c_str() << endl;
00520       prepareExit(criteria,destroyC,bstraps);
00521       return 5;
00522     }
00523     cout << "Read saved AdaBoost from file " << resumeFile.c_str()
00524          << " with " << ab.nTrained() << " trained classifiers." << endl;
00525   }
00526   if( skipInitialEventReweighting ) ab.skipInitialEventReweighting(true);
00527 
00528   // add trainable classifiers
00529   for( int i=0;i<useC.size();i++ ) {
00530     if( !ab.addTrainable(useC[i].first,useC[i].second) ) {
00531       cerr << "Unable to add classifier " << i << " of type " 
00532            << useC[i].first->name() << " to AdaBoost." << endl;
00533       prepareExit(criteria,destroyC,bstraps);
00534       return 6;
00535     }
00536   }
00537 
00538   // train
00539   if( !ab.train(verbose) )
00540     cerr << "AdaBoost terminated with error." << endl;
00541   if( ab.nTrained() == 0 ) {
00542     cerr << "Unable to train AdaBoost." << endl;
00543     prepareExit(criteria,destroyC,bstraps);
00544     return 7;
00545   }
00546   else {
00547     cout << "AdaBoost finished training with " << ab.nTrained() 
00548          << " classifiers." << endl;
00549   }
00550 
00551   // save trained AdaBoost
00552   if( !outFile.empty() ) {
00553     if( !ab.store(outFile.c_str()) ) {
00554       cerr << "Cannot store AdaBoost in file " << outFile.c_str() << endl;
00555       prepareExit(criteria,destroyC,bstraps);
00556       return 8;
00557     }
00558   }
00559 
00560   // save reweighted data
00561   if( !weightedDataOut.empty() ) {
00562     if( !ab.storeData(weightedDataOut.c_str()) ) {
00563       cerr << "Cannot store weighted AdaBoost data to file " 
00564            << weightedDataOut.c_str() << endl;
00565       prepareExit(criteria,destroyC,bstraps);
00566       return 9;
00567     }
00568   }
00569 
00570   // exit
00571   prepareExit(criteria,destroyC,bstraps);
00572   return 0;
00573 }

void prepareExit	(	vector< SprAbsTwoClassCriterion * > &	criteria,
		vector< SprAbsClassifier * > &	classifiers,
		vector< SprIntegerBootstrap * > &	bstraps
	)

Definition at line 80 of file SprBoosterApp.cc.

References i.

00083 {
00084   for( int i=0;i<criteria.size();i++ ) delete criteria[i];
00085   for( int i=0;i<classifiers.size();i++ ) delete classifiers[i];
00086   for( int i=0;i<bstraps.size();i++ ) delete bstraps[i];
00087 }


Functions
void	help (const char *prog)
int	main (int argc, char **argv)
void	prepareExit (vector< SprAbsTwoClassCriterion * > &criteria, vector< SprAbsClassifier * > &classifiers, vector< SprIntegerBootstrap * > &bstraps)

SprBoosterApp.cc File Reference

Functions

Function Documentation