main.cpp

/**
 * DESCRIPTION OF THE FILE
 *
 * @author Michal Kravčenko
 * @date 14.6.18 -
 */

#include <iostream>
#include <cstdio>
#include <fstream>
#include <vector>
#include <utility>
#include <algorithm>
#include <boost/archive/text_oarchive.hpp>
#include <boost/archive/text_iarchive.hpp>
#include "Network/NeuralNetwork.h"
#include "Neuron/NeuronLinear.h"
#include "Neuron/NeuronLogistic.h"
#include "NetConnection/Connection.h"
#include "NetConnection/ConnectionWeightIdentity.h"

#include "LearningMethods/ParticleSwarm.h"
#include "Neuron/NeuronBinary.h"
#include "Neuron/NeuronTanh.h"
#include "DataSet/DataSet.h"

//TODO rewrite "tests" to separate examples

//TODO prepsat tak, aby neuronova sit managovala destruktory vsech potrebnych objektu (kvuli serializaci)
/**
 * Test of simple neural network
 * Network should evaluate the function f(x) = x + 1
 */
void test1( ){
    std::vector<double> in(1);
    std::vector<double> out(1);

    NeuralNetwork net;
    NeuronLinear* u1 = new NeuronLinear(1.0, 1.0); //f(x) = x + 1.0
    NeuronLinear* u2 = new NeuronLinear(0.0, 1.0); //f(x) = x
    int idx1 = net.add_neuron(u1);
    int idx2 = net.add_neuron(u2);
////////////////////// SIMPLE EDGE WEIGHT ////////////////////////////////////////
//    net.add_connection_simple(idx1, idx2, -1, 1.0);
////////////////////// END SIMPLE EDGE WEIGHT ////////////////////////////////////////

/////////////////////////BEGIN OF COMPLEX EDGE WEIGHT//////////////////////////////
   std::function<double(double *, size_t*, size_t)> weight_function = [](double * weight_array, size_t * index_array, size_t n_params){
        //w(x, y) = x + y
        double a = weight_array[index_array[0]];
        double b = weight_array[index_array[1]];
//        printf("eval: %f, %f\n",  a, b);
        return (a + 0.0 * b);
    };
    size_t  weight_indices [2] = {0, 1};
    double weight_values [2] = {1.0, 5.0};
    net.add_connection_general(idx1, idx2, &weight_function, weight_indices, weight_values, 2);
/////////////////////////END OF COMPLEX EDGE WEIGHT//////////////////////////////
    for(int i = 0; i < 20; ++i){
        in[0] = 0.05 * i;
        net.eval_single(in, out);

        printf("x = %3.2f, f(x) = %3.2f, expected output = %3.2f\n", in[0], out[0], in[0] + 1.0);
    }

    //clean-up phase
    delete u1;
    delete u2;
}

/**
 * Test of DataSet serialization
 */
std::vector<double> out_f(std::vector<double> v) {
    double sum = 0;
    for(auto& e : v) {
        sum += e;
    }

    std::vector<double> out{sum*2, sum*3, sum*4};
    return out;
}

void test3() {

    /* Manually created data set */
    std::cout << "Manually created data set" << std::endl;
    std::vector<std::pair<std::vector<double>, std::vector<double>>> data_vec;
    std::vector<double> inp, out;

    for(int i = 0; i < 3; i++) {
        inp.push_back(i);
        out.push_back(i+4);
    }

    data_vec.emplace_back(std::make_pair(inp, out));

    DataSet ds(&data_vec);

    ds.print_data();

    ds.store_text("stored_data.4ds");

    DataSet ds2("stored_data.4ds");

    ds2.print_data();

    /* Isotropic data set */
    std::cout << "Isotropic data set:" << std::endl;
    std::vector<double> bounds{0,3,0,2};

    std::cout << "original one:" << std::endl;
    DataSet ds3(bounds, 5, out_f, 3);
    ds3.print_data();
    ds3.store_text("stored_data2.4ds");

    std::cout << "loaded one:" << std::endl;
    DataSet ds4("stored_data2.4ds");
    ds4.print_data();
}

double particle_swarm_test_function(double *x){
//    return x[0] * x[1] - x[0] * x[0] + x[1] * x[2];
    return x[0] * x[0];
}

NeuralNetwork net;
std::vector<std::vector<double>*> *train_data_input;
std::vector<std::vector<double>*> *train_data_output;

double test_particle_swarm_neural_net_error_function(double *weights){

    net.copy_weights(weights);

    unsigned int dim_out = train_data_output->at(0)->size();
//    unsigned int dim_in = train_data_input->at(0)->size();

    double error = 0.0, val;

    std::vector<double> output( dim_out );
    for(unsigned int i = 0; i < train_data_input->size(); ++i){

        net.eval_single(*train_data_input->at(i), output);

        for(unsigned int j = 0; j < dim_out; ++j){
            val = output[j] - train_data_output->at(i)->at(j);
            error += val * val;
        }

    }

    /*
    printf("INPUT: ");
    for(unsigned int i = 0; i < dim_in; ++i){
        printf("%f ", weights[i]);
    }
    printf(", ERROR: %f\n", 0.5 * error);
    */
    return 0.5 * error;
}

//TODO proper memory management
void test_particle_swarm_neural_net(){

//    unsigned int dim_in = 2;
//    unsigned int dim_out = 1;
//
//    /* TRAIN DATA DEFINITION */
//    train_data_input = new std::vector<std::vector<double>*>();
//    train_data_output = new std::vector<std::vector<double>*>();
//
//    std::vector<double> *input_01 = new std::vector<double>(dim_in);
//    std::vector<double> *input_02 = new std::vector<double>(dim_in);
//
//    std::vector<double> *output_01 = new std::vector<double>(dim_out);
//    std::vector<double> *output_02 = new std::vector<double>(dim_out);
//
//    (*input_01)[0] = 0.0;
//    (*input_01)[1] = 1.0;
//    (*output_01)[0] = 0.5;
//
//    (*input_02)[0] = 1.0;
//    (*input_02)[1] = 0.5;
//    (*output_02)[0] = 0.75;
//
//    train_data_input->push_back(input_01);
//    train_data_output->push_back(output_01);
//
//    train_data_input->push_back(input_02);
//    train_data_output->push_back(output_02);
//    /* NETWORK DEFINITION */
//
//
//    NeuronLinear* i1 = new NeuronLinear(0.0, 1.0); //f(x) = x
//    NeuronLinear* i2 = new NeuronLinear(0.0, 1.0); //f(x) = x
//
////    NeuronLogistic* o1 = new NeuronLogistic(1.0, 0.0); //f(x) = (1 + e^(-x + 0.0))^(1.0)
//    NeuronLinear* o1 = new NeuronLinear(1.0, 2.0); //f(x) = 2x + 1
//
//    int idx1 = net.add_neuron(i1);
//    int idx2 = net.add_neuron(i2);
//    int idx3 = net.add_neuron(o1);
//
//    net.add_connection_simple(idx1, idx3, -1, 1.0);
//    net.add_connection_simple(idx2, idx3, -1, 1.0);
//
//    /* PARTICLE SETUP */
//    double (*F)(double*) = &test_particle_swarm_neural_net_error_function;
//
//    unsigned int n_edges = 2;
//    unsigned int dim = n_edges, max_iters = 2000;
//
//
//    double domain_bounds [4] = {-800.0, 800.0, -800.0, 800.0};
//
//    double c1 = 0.5, c2 = 1.5, w = 0.8;
//
//    unsigned int n_particles = 10;
//
//    ParticleSwarm swarm_01(F, dim, domain_bounds, c1, c2, w, n_particles, max_iters);
//
//    swarm_01.optimize(0.5, 0.02);
//
//    /* CLEANUP PHASE */
//    for( std::vector<double> *input: *train_data_input){
//        delete input;
//    }
//    for( std::vector<double> *output: *train_data_output){
//        delete output;
//    }
//
//    delete train_data_output;
//    delete train_data_input;

}
//
//void test_particle_swarm(){
//    double (*F)(double*) = &particle_swarm_test_function;
//
//    unsigned int dim = 3, max_iters = 100;
//
////    double domain_bounds [2] = {2.0, 3.0};
//    double domain_bounds [6] = {-3.0, 3.0, 2.0, 5.0, 1.0, 15.0};
//
//    double c1 = 0.5, c2 = 1.5, w = 1.0;
//
//    unsigned int n_particles = 1000;
//
//    double accuracy = 1e-6;
//
//    ParticleSwarm swarm_01(F, dim, domain_bounds, c1, c2, w, n_particles, max_iters);
//
//    swarm_01.optimize(0.5, accuracy);
//}



int main(int argc, char** argv){

//    test1();

//    test_particle_swarm();

    test_particle_swarm_neural_net();

//    test2();

//    test3();

    return 0;
}