#include <stdio.h>
#include "MatDataSet.h"
#include "ClassFormatDataSet.h"
#include "ClassNLLCriterion.h"
#include "MSECriterion.h"
#include "OneHotClassFormat.h"
#include "ClassMeasurer.h"
#include "MSEMeasurer.h"

#include "StochasticGradient.h"
#include "KFold.h"

#include "ConnectedMachine.h"
#include "Linear.h"
#include "Tanh.h"
#include "LogSoftMax.h"

#include "MeanVarNorm.h"
#include "DiskXFile.h"
#include "CmdLine.h"
#include "Random.h"

using namespace Torch;


int main (void) {

	CmdLine cmd;
	Allocator *allocator = new Allocator;

	//===================== MLP =======================
    ConnectedMachine mlp;

    Linear *c1 = new(allocator) Linear(/*input*/ 2,/*output2Hidden*/ 4); //connectors
    //c1->setROption("weight decay", 0.5);
    
	Tanh *c2 = new(allocator) Tanh(/*hiddenSize*/ 4); //hidden layer
    
	Linear *c3 = new(allocator) Linear(/*hiddenSize*/ 4,/*outputSize*/ 1); //connectors
    //c3->setROption("weight decay", 0.5);

	//LogSoftMax *c4 = new(allocator) LogSoftMax(/*OutputSize*/ 1); //RegressionMode

    mlp.addFCL(c1);    
    mlp.addFCL(c2);
    mlp.addFCL(c3);
   // mlp.addFCL(c4);

	 // Initialize the MLP
	mlp.build();
	mlp.setPartialBackprop();
	printf("MLP initialized\n");

    //====================== DataSet ====================================
	MatDataSet *data = new(allocator) MatDataSet("XOR-ProblemData.txt", 2, 1);
	printf("Data loaded\n");
	
	//====================== Measure ====================================
	  // The list of measurers...
	MeasurerList measurers;
	DataSet *valid_data = new(allocator) MatDataSet("XOR-ProblemData.txt", 2, 1);
	DiskXFile mse_file("the_mse_error", "w");

	MSEMeasurer *valid_mse_meas = new(allocator) MSEMeasurer(mlp.outputs, valid_data, &mse_file);
	measurers.addNode(valid_mse_meas);    
	printf("Measure initialized\n");

	//====================== Trainer ====================================
	Criterion *criterion = NULL;
	criterion = new(allocator) MSECriterion(/*outputSize*/ 1);

	StochasticGradient trainer(&mlp, criterion);

	//trainer.setIOption("max iter", 100);
	trainer.setROption("end accuracy", 0.0001);
	trainer.setROption("learning rate", 0.1);
	//trainer.setROption("learning rate decay", 0.1);


	//train data
	trainer.test(&measurers);

	trainer.train(data, &measurers);

	trainer.test(&measurers);

	delete allocator;

	printf("The End\n");
    return 0;
}