Simple-Neural-Network-Training-Example

This repository contains a simple neural network implementation in Python. The neural network uses a sigmoid activation function and adjusts weights through training iterations.

Installation

No special installation is required for this project. Just make sure you have Python installed.

Usage

To run the program, simply execute the nn_example.py script.

python nn_example.py

Code Explanation

from math import exp
from random import random as r

def s(x):
    return 1 / (1 + exp(-x))

i = [1, 0, 1]

w = []
for _ in range(3):
    w = w + [r()]

print("Random initial weights:")
print(w)

to = 1  

for _ in range(20000):
    il = i

    o = 0
    for j in range(3):
        o = o + il[j] * w[j]

    o = s(o)   

    err = to - o

    adj = [0] * 3

    for j in range(3):
        adj[j] = il[j] * (err * (o * (1 - o)))
        w[j] = w[j] + adj[j]

print("Weights after training:")
print(w)


o = 0
for j in range(3):
    o = o + i[j] * w[j]

o = s(o) 

print("Result:")
print(o)

How It Works

1. Initialization:

   • The program begins by defining a sigmoid activation function s.
   • An input list i and an initial weight list w with random values are created.

2. Training:

   • The program trains the neural network over 20,000 iterations.
   • In each iteration, the output is calculated by summing the product of inputs and weights, then passing through the sigmoid function.
   • The error is computed as the difference between the target output (to) and the actual output.
   • The weights are adjusted based on the error using gradient descent.

3. Result:

   • After training, the adjusted weights are printed.
   • The program then computes the output using the trained weights and prints the final result.

First Attempt

This is a simple implementation to demonstrate the basic principles of neural networks. Improvements and enhancements can be made for more complex tasks.