About

The following repository is a simple implementation for Neural Network using the C programming language.

How to use

The Neural Network functions are stated in the header file ann.h, each network is associated with a structure defined:

typedef struct networks{
    int n_layers;
    int dim[LAYER_SIZE];
    double weights[LAYER_SIZE][MAX_SIZE][MAX_SIZE];
    double biases[LAYER_SIZE][MAX_SIZE];
}network;

Macros

MAX_SIZE - The Maximum possible size for the number of neurons in a single layer ( this is used to determine the maximum possible dimensions of weights and biases)

LAYER_SIZE - Maximum possible number of layers.

Note - Setting high values for MAX_SIZE and LAYER_SIZE might take up too much memory and your program might be unable to run.

Initializing

It is advisable to initialize the network structure and allocate memory for it in the heap using malloc function

#include <stdio.h>
#include <stdlib.h>
#include "ann.h"

network* ann = (network *)malloc(sizeof(network));

Intializing 2D and 3D arrays in the heap can be done using the alloc.h header provided in this repository

#include <stdio.h>
#include <stdlib.h>
#include "ann.h"
#include "alloc.h"

double **train_data = init_2Darray(3,2); //Equivalent to double train_data[3][2]; but allocated in the heap 
double **test_data = init_3Darray(3,4); //Equivalent to double test_data[3][4]; but allocated in the heap

Functions

init_ann(network *,int dimensions[],no_of_layers) - Automatically Initializes the neural network and assigns initial (random) values for weights and biases based on the network structure defined in the dimensions array.

init_ann_with_weights(network *,int dimensions[],double weights[][][], double biases[][],no_of_layers) - Initializes neural network with weights and bias input as an argument.

train(network *,double **data,length,learning_rate) - Trains the network using gradient descent back-propogation algorithm and data provided.

test(network *,double **data,length) - Tests the network based on the test data input and shows accuracy.

predict(network* , double *data) - Gives output prediction for a particular test case stated in the data input.

Examples

Initializing a neural network with 3 layers ( 2 neurons in input, 2 neurons in the hidden layers, 1 neuron in the output layer ):

#include <stdio.h>
#include <math.h>
#include <stdlib.h>
#include "ann.h"

int main(){
  
    network *ann = (network*)malloc(sizeof(network));
    int dim[3] = {2,2,1}; //Dimension array
    init_ann(ann,dim,3); // Automatically assigns random values for weights and biases and initializes network 
    
    return 0;
}

1. Neural network for predicting XOR - XOR network

2. Neural network for classifying handwritten digits based on kaggle dataset - digit classifier