logistic_regression_SAG.cpp

Go to the documentation of this file.

#include <shark/Data/SparseData.h>
#include <shark/ObjectiveFunctions/Loss/CrossEntropy.h>
#include <shark/Algorithms/Trainers/LinearSAGTrainer.h>
 
#include <shark/Core/Timer.h>
#include <iostream>
using namespace shark;
using namespace std;
 
 
template<class InputType>
void run(LabeledData<InputType,unsigned int> const& data, double alpha, unsigned int epochs){
    CrossEntropy<unsigned int, RealVector> loss;
    LinearClassifier<InputType> model;
    
    
    LinearSAGTrainer<InputType,unsigned int> trainer(&loss,alpha);
    trainer.setEpochs(epochs);
    
    Timer time;
    trainer.train(model, data);
    double time_taken = time.stop();
    
    cout << "Cross-Entropy: " << loss(data.labels(),model.decisionFunction()(data.inputs()))<<std::endl;
    cout << "Time:\n" << time_taken << endl;
}
int main(int argc, char **argv) {
    ClassificationDataset data_dense;
    importSparseData(data_dense, "mnist",0,8192);
    data_dense = transformLabels(data_dense, [](unsigned int y){ return y%2;});
    LabeledData<CompressedRealVector,unsigned int> data_sparse;
    importSparseData(data_sparse, "rcv1_train.binary",0,8192);
    
    double alpha = 0.1;
    run(data_dense, alpha, 200);
    run(data_sparse, alpha, 2000);
    
}