NICE_VisLearning/progs/testImageNetBinaryGPBaseline.cpp

/** 
* @file testImageNetBinaryGPBaseline.cpp
* @brief perform ImageNet tests with binary tasks for OCC using the baseline GP
* @author Alexander Lütz
* @date 29-05-2012 (dd-mm-yyyy)

*/
#include "core/basics/Config.h"

#ifdef NICE_USELIB_MATIO

#include "core/basics/Timer.h"
#include "core/vector/SparseVectorT.h"
#include "core/algebra/CholeskyRobust.h"
#include "core/vector/Algorithms.h"

#include "vislearning/cbaselib/ClassificationResults.h"
#include "vislearning/baselib/ProgressBar.h"

#include "core/matlabAccess/MatFileIO.h"
#include "vislearning/matlabAccessHighLevel/ImageNetData.h"

// #include "fast-hik/tools.h"

using namespace std;
using namespace NICE;
using namespace OBJREC;

double measureDistance ( const NICE::SparseVector & a, const NICE::SparseVector & b, const double & sigma = 2.0)//, const bool & verbose = false)
{
  double inner_sum(0.0);

  double d;    
    
  //new version, where we needed on average 0.001707 s for each test sample
  NICE::SparseVector::const_iterator aIt = a.begin();
  NICE::SparseVector::const_iterator bIt = b.begin();
   
  while ( (aIt != a.end()) && (bIt != b.end()) )
  {
    if (aIt->first == bIt->first)
    {
      d = ( aIt->second - bIt->second );      
      inner_sum += d * d;
      aIt++;
      bIt++;
    }
    else if ( aIt->first < bIt->first)
    {
      inner_sum += aIt->second * aIt->second;
      aIt++;      
    }
    else
    {
      inner_sum += bIt->second * bIt->second;
      bIt++;       
    }
  }
  
  //compute remaining values, if b reached the end but not a
  while (aIt != a.end())
  {
    inner_sum += aIt->second * aIt->second;
    aIt++; 
  }
  //compute remaining values, if a reached the end but not b
  while (bIt != b.end())
  {
    inner_sum += bIt->second * bIt->second;
    bIt++; 
  }  
  inner_sum /= (2.0*sigma*sigma);
  
  return exp(-inner_sum); //expValue;
}

void readParameters(const string & filename, const int & size, NICE::Vector & parameterVector)
{
  parameterVector.resize(size);
  parameterVector.set(0.0);
  
  ifstream is(filename.c_str());
  if ( !is.good() )
    fthrow(IOException, "Unable to read parameters.");  
//
  string tmp;
  int cnt(0);
  while (! is.eof())
  {
    is >> tmp;
    parameterVector[cnt] = atof(tmp.c_str());
    cnt++;
  }
//   
  is.close(); 
}


/** 
    test the basic functionality of fast-hik hyperparameter optimization 
*/
int main (int argc, char **argv)
{   
  std::set_terminate(__gnu_cxx::__verbose_terminate_handler);

  Config conf ( argc, argv );
  string resultsfile = conf.gS("main", "results", "results.txt" );
  double kernelSigma = conf.gD("main", "kernelSigma", 2.0);
  int nrOfExamplesPerClass = conf.gI("main", "nrOfExamplesPerClass", 50);
  nrOfExamplesPerClass = std::min(nrOfExamplesPerClass, 100); // we do not have more than 100 examples per class
  int nrOfClassesToConcidere = conf.gI("main", "nrOfClassesToConcidere", 1000);
  nrOfClassesToConcidere = std::min(nrOfClassesToConcidere, 1000); //we do not have more than 1000 classes

  string sigmaFile = conf.gS("main", "sigmaFile", "approxVarSigma.txt");  
  string noiseFile = conf.gS("main", "noiseFile", "approxVarNoise.txt");  
  
  
  NICE::Vector sigmaParas(nrOfClassesToConcidere,kernelSigma);
  NICE::Vector noiseParas(nrOfClassesToConcidere,0.0);
  
  readParameters(sigmaFile,nrOfClassesToConcidere, sigmaParas);
  readParameters(noiseFile,nrOfClassesToConcidere, noiseParas);
  
  std::vector<SparseVector> trainingData;
  NICE::Vector y;
  
  std::cerr << "Reading ImageNet data ..." << std::endl;
  bool imageNetLocal = conf.gB("main", "imageNetLocal" , false);
  string imageNetPath;
  if (imageNetLocal)
    imageNetPath = "/users2/rodner/data/imagenet/devkit-1.0/";
  else
    imageNetPath = "/home/dbv/bilder/imagenet/devkit-1.0/";

  ImageNetData imageNetTrain ( imageNetPath + "demo/" );

  imageNetTrain.preloadData( "train", "training" );
  trainingData = imageNetTrain.getPreloadedData();
  y = imageNetTrain.getPreloadedLabels();
    
  std::cerr << "Reading of training data finished" << std::endl;
  std::cerr << "trainingData.size(): " << trainingData.size() << std::endl;
  std::cerr << "y.size(): " << y.size() << std::endl;
  
  std::cerr << "Reading ImageNet test data files (takes some seconds)..." << std::endl;
  ImageNetData imageNetTest ( imageNetPath + "demo/" );
  imageNetTest.preloadData ( "val", "testing" );
  imageNetTest.loadExternalLabels ( imageNetPath + "data/ILSVRC2010_validation_ground_truth.txt" );  
  
  double OverallPerformance(0.0);  
  
  for (int cl = 0; cl < nrOfClassesToConcidere; cl++)
  {
    std::cerr << "run for class " << cl << std::endl;
    int positiveClass = cl+1;
    // ------------------------------ TRAINING ------------------------------
  
    kernelSigma = sigmaParas[cl];
    
    std::cerr << "using sigma: " << kernelSigma << " and noise " << noiseParas[cl] << std::endl;
    Timer tTrain;
    tTrain.start();
    NICE::Matrix kernelMatrix (nrOfExamplesPerClass, nrOfExamplesPerClass, 0.0);
      
    //now compute the kernelScores for every element
    double kernelScore(0.0);
    for (int i = cl*100; i < cl*100+nrOfExamplesPerClass; i++)
    {
      for (int j = i; j < cl*100+nrOfExamplesPerClass; j++)
      {
        kernelScore = measureDistance(trainingData[i],trainingData[j], kernelSigma);//optimalParameters[cl]);
        kernelMatrix(i-cl*100,j-cl*100) = kernelScore;
        if (i != j)
          kernelMatrix(j-cl*100,i-cl*100) = kernelScore;
      }
    }  
    
    //adding some noise, if necessary
    if (noiseParas[cl] != 0.0)
    {
      kernelMatrix.addIdentity(noiseParas[cl]);
    }
    else
    {
      //zero was already set
    }    
   
    //compute its inverse
    //noise is already added :)
    Timer tTrainPrecise;
    tTrainPrecise.start();     
    
    CholeskyRobust cr  ( false /* verbose*/, 0.0 /*noiseStep*/, false /* useCuda*/);
    
    NICE::Matrix choleskyMatrix (nrOfExamplesPerClass, nrOfExamplesPerClass, 0.0);      
    cr.robustChol ( kernelMatrix, choleskyMatrix );    
    
    tTrainPrecise.stop(); 
    std::cerr << "Precise time used for training class " << cl << ": " << tTrainPrecise.getLast() << std::endl;    
    
    tTrain.stop();
    std::cerr << "Time used for training class " << cl << ": " << tTrain.getLast() << std::endl;    
       
    std::cerr << "training done - now perform the evaluation" << std::endl;


    // ------------------------------ TESTING ------------------------------
   
    ClassificationResults results;
    std::cerr << "Classification step ... with " << imageNetTest.getNumPreloadedExamples() << " examples" << std::endl;
    ProgressBar pb;
    Timer tTest;
    tTest.start();    
    Timer tTestSingle;
    double timeForSingleExamples(0.0);
    for ( uint i = 0 ; i < (uint)imageNetTest.getNumPreloadedExamples(); i++ )
    {
      pb.update ( imageNetTest.getNumPreloadedExamples() );

      //get the precomputed features
      const SparseVector & svec = imageNetTest.getPreloadedExample ( i );
      
      //compute (self-)similarities
      double kernelSelf (measureDistance(svec,svec, kernelSigma) );
      NICE::Vector kernelVector (nrOfExamplesPerClass, 0.0);
      
      for (int j = 0; j < nrOfExamplesPerClass; j++)
      {
        kernelVector[j] = measureDistance(trainingData[j+cl*100],svec, kernelSigma);
      }     
      
      //compute the resulting score
      tTestSingle.start();
      NICE::Vector rightPart (nrOfExamplesPerClass);
      choleskySolveLargeScale ( choleskyMatrix, kernelVector, rightPart );
        
      double uncertainty = kernelSelf - kernelVector.scalarProduct ( rightPart );
      tTestSingle.stop();
      timeForSingleExamples += tTestSingle.getLast();
      
      //this is the standard score-object needed for the evaluation
      FullVector scores ( 2 );
      scores[0] = 0.0;
      scores[1] = 1.0 - uncertainty;

      ClassificationResult r ( scores[1]<0.5 ? 0 : 1, scores );    
      
      // set ground truth label
      r.classno_groundtruth = (((int)imageNetTest.getPreloadedLabel ( i )) == positiveClass) ? 1 : 0;
      
      //we could write the resulting score on the command line
//       std::cerr << "scores: " << std::endl;
//       scores >> std::cerr;
      //as well as the ground truth label
//       std::cerr << "gt: " <<  r.classno_groundtruth << " -- " << r.classno << std::endl;
      
      results.push_back ( r );
    }
    
    tTest.stop();
    std::cerr << "Time used for evaluating class " << cl << ": " << tTest.getLast() << std::endl;       
    
    timeForSingleExamples/= imageNetTest.getNumPreloadedExamples();
    std::cerr << "Time used for evaluation single elements of class " << cl << " : " << timeForSingleExamples << std::endl;
    

    // we could also write the results to an external file. Note, that this file will be overwritten in every iteration
    // so if you want to store all results, you should add a suffix with the class number
//     std::cerr << "Writing results to " << resultsfile << std::endl;
//     results.writeWEKA ( resultsfile, 1 );
    double perfvalue = results.getBinaryClassPerformance( ClassificationResults::PERF_AUC );

    std::cerr << "Performance: " << perfvalue << std::endl;
    
    OverallPerformance += perfvalue;    
  }
  
  OverallPerformance /= nrOfClassesToConcidere;
  
  std::cerr << "overall performance: " << OverallPerformance << std::endl;
  
  return 0;
}

#else
int main (int argc, char **argv)
{
  
}

#endif