Overview

A minimalistic and customizable neural network built from scratch using Python and NumPy.

🧠 Features

• Modular Design:

  • Organized class structure for network components.
  • Easily extendable to add new layers, activation functions, and loss functions.

• Customizable Components:

  • Fully connected (dense) layers with adjustable input and output sizes.
  • Multiple activation and loss functions that can be modified.

• Model Engine

  • An "engine" that trains, evaluates, saves, and loads the neural network.
  • Adjustable epochs and learning rates for training.
  • Custom accuracy functions for evaluation.

💡 Performance

• Trained on the MNIST dataset, achieving 95% accuracy on the testing set with custom hyperparameters.

🛠️ Requirements

• Create a virtual environment and install dependencies:

  pip install -r requirements.txt

---

For a specific usage example, check out MNIST_model_train.py and MNIST_model_predict.py.