Basic neural network (Perceptron)

Data de publicació: Mar 21, 2017 9:58:4 PM

//Dedicated to Llorenç Magraner

#include <stdio.h>

#include <stdlib.h>

#include <math.h>

#include <time.h>

//Learning rate and Threshold can be changed it depends on the problem

#define LR    0.1

#define MAX_EPOCHS     5000

float randomFloat(){

    return (float)rand() / (float)RAND_MAX;

}

int _compute_Output(float W[], float x, float y){

    return (W[0] + x * W[1] + y * W[2] >= 0 ) ? 1 : -1;

}

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

    srand(time(NULL));

    float W[3];

    float x[300], y[300];

    int outputs[300];

    float localError, globalError;

    int numberTests, i, p, epoch, output_calc;

    FILE *fp;

    if ((fp = fopen("test_1Q3Q.txt", "r")) == NULL) {printf("No datafile found.\n");exit(1);}

    i = 0;

    while (fscanf(fp, "%f %f %d", &x[i], &y[i], &outputs[i]) != EOF) i++;

    numberTests = i;

     printf("Training number %d \n",numberTests); getchar();

     //Bias, threshold

    W[0] = randomFloat();

    //Weights

    W[1] = randomFloat();

    W[2] = randomFloat();

    epoch = 0;

    do {

        epoch++;

        globalError = 0;

        for (p = 0; p < numberTests; p++) {

            output_calc = _compute_Output(W, x[p], y[p]);

            localError = outputs[p] - output_calc;

            //Bias

            W[0] += LR * localError;

            //Weights

            W[1] += LR * localError * x[p];

            W[2] += LR * localError * y[p];

            globalError += ( localError * localError);

        }

        globalError =  globalError / numberTests ;

        printf("Epoca %d: GlobalError: %.4f sqrt(GlobalError) = %.4f\n", epoch, globalError, sqrt(globalError));

    } while ( globalError != 0 && epoch <= MAX_EPOCHS );

     printf("\nGrading Line: %.2f*x + %.2f*y + %.2f = 0\n", W[1], W[2], W[0]);

    return 0;

}

gcc -lm perceptron.c -o perceptron

Perceptron

Great sequence of seven videos explaining what a neural network is and how it works.