/*
 * main.cpp
 *
 *  Created on: 09.06.2011
 *      Author: sven
 */

#include "MLPConfig.h"
#include "MLP.h"

#include <boost/foreach.hpp>
#include <boost/program_options.hpp>
#include <boost/tokenizer.hpp>
#include <string>
#include <vector>
#include <iostream>
#include <cstdlib>
#include <cmath>
#include <csignal>

namespace po = boost::program_options;

const char* MLP_XOR_CONFIGFILE = "Init_MLP.txt";
const char* MLP_PCA_TRAINFILE = "train_pca";
const char* MLP_PCA_TESTFILE = "test_pca";
const char* MLP_RAW_TRAINFILE = "train_raw";
const char* MLP_RAW_TESTFILE = "test_raw";
const char* MLP_PCA_CONFIGFILE = "mlp_pca_config.txt";
const char* MLP_RAW_CONFIGFILE = "mlp_raw_config.txt";

const uint32_t VALIDATION_DATA_RATIO = 20; /* every x th value will become a validition pattern */
const double EPSILON = 0.001;

void mlp_xor(MLPConfig& config);
void mlp_train_pca(MLPConfig& config);
void mlp_train_raw(MLPConfig& config);
void mlp_test_pca(MLPConfig& config);
void mlp_test_raw(MLPConfig& config);
void parseTrainData(const char*,Traindata&,Traindata&);
void signal_handler(int signal);
static MLP* mlpPtr = NULL;

void signal_handler(int signal)
{
    std::cout << "Terminating..." << std::endl;
    if (mlpPtr) {
        mlpPtr->stop();
    }
    mlpPtr = NULL;
}

double toDouble(const std::string& t)
{
	double res;
	std::istringstream ss(t);
	ss >> res;
	return res;
}

void parseTrainData(const char* filename, Traindata& td, Traindata& vd)
{
	typedef boost::char_separator<char> TrainSepChar;
	typedef tokenizer<TrainSepChar> TrainTokenizer;
	TrainSepChar sepCol(":");
	TrainSepChar sepSpace(" ");
	ifstream inFile(filename);
	if (!inFile.is_open()) {
		std::cerr << "error opening data file " << filename << std::endl;
		exit(1);
	}
	std::string line;
	uint32_t counter = 0;
	while(inFile.good())
	{
		getline(inFile,line);
		if (line.empty()) {
			continue;
		}
		TrainTokenizer lineTok(line,sepCol);
		std::vector<std::string> lineToks;
		BOOST_FOREACH(string t, lineTok)
		{
			lineToks.push_back(t);
		}
		Trainingpair tp;
		TrainTokenizer targetTok(lineToks.at(0),sepSpace);
		BOOST_FOREACH(string t,targetTok)
		{
			tp.mTarget.push_back(toDouble(t));
		}
		TrainTokenizer patternTok(lineToks.at(1),sepSpace);
		BOOST_FOREACH(string t,patternTok)
		{
			if (t.find(";") != std::string::npos) {
				continue;
			}
			tp.mPattern.push_back(toDouble(t));
		}
		if ( (counter % VALIDATION_DATA_RATIO) != 0 ) {
			td.push_back(tp);
		} else {
			vd.push_back(tp);
		}
		counter++;
	}
	inFile.close();
}

void mlp_train_pca(MLPConfig& config)
{
	config.setNumInputNeurons(90U);
	config.setNumHiddenNeurons(90U);
	config.setNumOutputNeurons(2U);
	config.initWeights(true);
	MLP mlp(config);
    mlpPtr = &mlp;
	Traindata td;
	Traindata vd;
	parseTrainData(MLP_PCA_TRAINFILE,td,vd);
	mlp.train(td,vd,config.getNumTrainingCycles(),MLP_PCA_CONFIGFILE);
}

void mlp_test_pca(MLPConfig& config)
{
	config.parseConfigfile(MLP_PCA_CONFIGFILE);
	if (!config.isValid()) {
		std::cerr << "error parsing config file " << MLP_PCA_CONFIGFILE  << std::endl;
		exit(1);
	}
	MLP mlp(config);
	Traindata td;
	Output output;
	parseTrainData(MLP_PCA_TESTFILE,td,td);
	uint32_t c;
	uint32_t num = 1;
	uint32_t errorCount = 0;
	double out,target;
	double res[2];
	BOOST_FOREACH(Trainingpair tp, td) {
		output.clear();
		mlp.propagate(tp.mPattern,output);
		bool errorFound = false;
		for (c = 0; c < 2; c++) {
			out = output.at(c);
			if (out >= 0.5) {
				res[c] = 1.0;
			} else {
				res[c] = 0.0;
			}
		}
		for (c = 0; c < 2; c++) {
			target = tp.mTarget.at(c);
			if (std::abs(res[c] - target) > EPSILON) {
				cout << num << " PCA Testerror: " << res[c] << " != " << target << endl;
				errorFound = true;
			} else {
				cout << num << " PCA OK: " << res[c] << " == " << target << endl;
			}
		}
		if (errorFound) {
			errorCount++;
		}
		num++;
	}
	cout << "PCA: " << errorCount << " errors in " << td.size() << " testpatterns" << endl;
}

void mlp_train_raw(MLPConfig& config)
{
	config.setNumInputNeurons(6U*151U);
	config.setNumHiddenNeurons(6U*151U);
	config.setNumOutputNeurons(2U);
	config.initWeights(true);
	MLP mlp(config);
    mlpPtr = &mlp;
	Traindata td;
	Traindata vd;
	parseTrainData(MLP_RAW_TRAINFILE,td,vd);
	mlp.train(td,vd,config.getNumTrainingCycles(),MLP_RAW_CONFIGFILE);
}

void mlp_test_raw(MLPConfig& config)
{
	config.parseConfigfile(MLP_RAW_CONFIGFILE);
	if (!config.isValid()) {
		std::cerr << "error parsing config file " << MLP_RAW_CONFIGFILE  << std::endl;
		exit(1);
	}
	MLP mlp(config);
	Traindata td;
	Output output;
	parseTrainData(MLP_RAW_TESTFILE,td,td);
	uint32_t c;
	uint32_t num = 1;
	uint32_t errorCount = 0;
	double out,target;
	double res[2];
	BOOST_FOREACH(Trainingpair tp, td) {
		output.clear();
		mlp.propagate(tp.mPattern,output);
		bool errorFound = false;
		for (c = 0; c < 2; c++) {
			out = output.at(c);
			if (out >= 0.5) {
				res[c] = 1.0;
			} else {
				res[c] = 0.0;
			}
		}
		for (c = 0; c < 2; c++) {
			target = tp.mTarget.at(c);
			if (std::abs(res[c] - target) > EPSILON) {
				cout << num << " RAW Testerror: " << res[c] << " != " << target << endl;
				errorFound = true;
			} else {
				cout << num << " RAW OK: " << res[c] << " == " << target << endl;
			}
		}
		if (errorFound) {
			errorCount++;
		}
		num++;
	}
	cout << "RAW: " << errorCount << " errors in " << td.size() << " testpatterns" << endl;
}

void mlp_xor(MLPConfig& config) {
	MLP mlp(config);
	uint8_t numPatterns = 4;
	Traindata td;
    const double patterns[][2] = {
    		{0.0 , 0.0},
    		{0.0 , 1.0},
    		{1.0 , 0.0},
    		{1.0 , 1.0}
	};
    const double targets[][1] = {
    		{0},
    		{1},
    		{1},
    		{0}
    };
    
    for (uint8_t i = 0; i < numPatterns ; i++) {
    	Trainingpair tp;
    	tp.mPattern.assign(patterns[i],patterns[i]+2);
    	tp.mTarget.assign(targets[i],targets[i]+1);
    	td.push_back(tp);
    }
    
    mlp.train(td,td,config.getNumTrainingCycles(),"mlp_xor_weights.txt");

    // test with other data (Aufgabe 3)
    int32_t i,j;
    int32_t minval = -40;
    int32_t maxval = 50;
    double x,y;
    Output result;
    Pattern test_pat;
    ofstream csvFile;
    const char* sep = " ";
    csvFile.open("mlp_xor_out.dat",ios::out);
    for (i = minval ; i < maxval ; i++)
    {
        x = ((double) i) / 10.0;
        for (j = minval ; j < maxval ; j++) {
            y = ((double) j) / 10.0;
            test_pat.clear();
            test_pat.push_back(x);
            test_pat.push_back(y);
            result.clear();
            mlp.propagate(test_pat,result);
            csvFile << x << sep << y;
            BOOST_FOREACH(double z,result)
            {
                 csvFile << sep << z;
            }
            csvFile << endl;
        }
        csvFile << endl;
    }
    csvFile.close();
    system("gnuplot -persist mlp_xor.plt");
}

int main(int argc, char* argv[])
{
    signal(SIGINT,&signal_handler);

	MLPConfig config;

	enum APP {
		APP_XOR,APP_TRAIN_RAW,APP_TRAIN_PCA,APP_TEST_RAW,APP_TEST_PCA
	};

	// default settings
	config.setLernrate(0.8);
	config.setUpdateMode(MLPConfig::UPDATE_MODE_BATCH);
	config.setMomentum(0.9);
    config.setErrorThreshold(0.01);
    config.setConfigAcceptanceErrorThreshold(0.1);
    config.setNumTrainingCycles(20000U);
	APP app = APP_XOR;

    try {
        po::options_description desc("Allowed options");
        desc.add_options()
            ("help", "produce help message")
            ("trainingcycles", po::value<uint32_t>(), "set number of training cycles. default 20000")
            ("lernrate", po::value<double>(), "set lernrate. default 0.8")
            ("momentum", po::value<double>(), "set momentum. default 0.9")
            ("maxerror", po::value<double>(), "set maximum error that stops training. default 0.01")
            ("acceptanceerror", po::value<double>(), "set error after which the configuration is accepted if error increases. default 0.1")
            ("updatemode", po::value<std::string>(), "set update mode <\"batch\"|\"single\">. default batch")
            ("app", po::value<std::string>(), "application <\"xor\"|\"trainraw\"|\"trainpca\"|\"testraw\"|\"testpca\">. default xor")
        ;

        po::variables_map vm;        
        po::store(po::parse_command_line(argc, argv, desc), vm);
        po::notify(vm);    

        if (vm.count("help")) {
            cout << desc << "\n";
            return 1;
        }
        if (vm.count("trainingcycles")) {
            uint32_t trainingcycles = vm["trainingcycles"].as<uint32_t>();
            config.setNumTrainingCycles(trainingcycles);
        }
        if (vm.count("lernrate")) {
            double lernrate = vm["lernrate"].as<double>();
            config.setLernrate(lernrate);
        }
        if (vm.count("momentum")) {
            double momentum = vm["momentum"].as<double>();
            config.setMomentum(momentum);
        }
        if (vm.count("maxerror")) {
            double maxerror = vm["maxerror"].as<double>();
            config.setErrorThreshold(maxerror);
        }
        if (vm.count("momentum")) {
            double acceptanceerror = vm["acceptanceerror"].as<double>();
            config.setConfigAcceptanceErrorThreshold(acceptanceerror);
        }
        if (vm.count("updatemode")) {
            std::string updatemode = vm["updatemode"].as<std::string>();
            if (updatemode.find("batch") != std::string::npos) {
                config.setUpdateMode(MLPConfig::UPDATE_MODE_BATCH);
            } else if(updatemode.find("single") != std::string::npos) {
                config.setUpdateMode(MLPConfig::UPDATE_MODE_SINGLE);
            }
        }
        if (vm.count("app")) {
        	std::string appOpt = vm["app"].as<std::string>();
        	if (appOpt.find("xor") != std::string::npos) {
        		app = APP_XOR;
        	} else if (appOpt.find("trainraw") != std::string::npos) {
        		app = APP_TRAIN_RAW;
        	} else if (appOpt.find("trainpca") != std::string::npos) {
        		app = APP_TRAIN_PCA;
        	} else if (appOpt.find("testraw") != std::string::npos) {
        		app = APP_TEST_RAW;
        	} else if (appOpt.find("testpca") != std::string::npos) {
        		app = APP_TEST_PCA;
        	}
        }
    }
    catch(std::exception& e) {
        cerr << "error: " << e.what() << "\n";
        return 1;
    }
    catch(...) {
        cerr << "Exception of unknown type!\n";
    }
    switch(app) {
    case APP_XOR:
    	config.parseConfigfile(MLP_XOR_CONFIGFILE);
		if (!config.isValid()) {
			std::cerr << "error parsing config file " << MLP_XOR_CONFIGFILE  << std::endl;
			exit(1);
		}
    	mlp_xor(config);
    	break;
    case APP_TRAIN_PCA:
    	mlp_train_pca(config);
    	break;
    case APP_TRAIN_RAW:
    	mlp_train_raw(config);
    	break;
    case APP_TEST_PCA:
    	mlp_test_pca(config);
    	break;
    case APP_TEST_RAW:
    	mlp_test_raw(config);
    	break;
    default:
    	break;
    }
	return 0;
}