A small neural network written in pure C++ that learns the XOR function in real-time using a single hidden layer. This project demonstrates a basic fully-connected feedforward network, backpropagation, and weight initialization using Xavier method. ANSI escape codes are used for colored console output.

Fully configurable via console input:

• Random or manual seed
• Number of epochs
• Learning rate
• Number of hidden neurons
• Display interval for training progress

Real-time training feedback:

• Epoch number
• XOR inputs and predicted output
• Loss per batch
• Tracks best loss and corresponding epoch
• Optionally displays final weights of the network
• ANSI-colored output for better readability

1 Clone the repository:

git clone https://github.com/Vladyslaa/SimpleNeuralNetwork.git
cd SimpleNeuralNetwork

2 Compile with GCC (or any C++ compiler, requires C++17 or later):

g++ -std=c++17 src/SimpleNeuralNetwork.cpp -O3 -o SimpleNeuralNetwork.exe 

3 Run the executable:

./SimpleNeuralNetwork.exe

4 Enter the requested configuration parameters

5 Observe the network learning XOR in real-time

Epoch 10000000 | 0 XOR 0 = 0.00000089 (logit: -13.92975366, target: 0)
Epoch 10000000 | 1 XOR 0 = 0.99999803 (logit: 13.13961680, target: 1)
Epoch 10000000 | 0 XOR 1 = 0.99999857 (logit: 13.45552810, target: 1)
Epoch 10000000 | 1 XOR 1 = 0.00000210 (logit: -13.07367795, target: 0)
  Loss: 0.00000160

----------------------------------------
Neural Network Training Complete!
----------------------------------------
Best Loss: 0.00000160 at Epoch 10000000

Final XOR Evaluation:
   0 XOR 0 = 0.00000089
   1 XOR 0 = 0.99999803
   0 XOR 1 = 0.99999857
   1 XOR 1 = 0.00000210

Would you like to see final weights? (y/n):

Console must support ANSI escape codes for color output.