Micrograd-Mini is a tiny scalar-valued autograd engine with a basic neural network framework — built from scratch, inspired by Andrej Karpathy's micrograd.

It supports reverse-mode automatic differentiation (backpropagation) and is capable of training a simple neural network (MLP) on tasks like XOR. Great for learning the internals of backprop!

• ✅ Reverse-mode autodiff (backpropagation)
• ✅ Dynamic computation graph (DAG)
• ✅ Custom Value class with gradients
• ✅ Tanh and ReLU activations
• ✅ Basic Neuron, Layer, and MLP implementation
• ✅ Trains on XOR dataset

micrograd-mini/
    ├── engine.py # Core autodiff engine (Value class)
    ├── nn.py # Neural network components
    ├── train.py # Training loop for XOR
    ├── example.py # Better example using XOR dataset 
└── README.md # Project info

• Python 3.7+
• No external libraries needed (pure Python)

python train.py

--- Final Predictions after Training ---

Input: [0.0, 0.0] => Predicted: 0.01 | Target: 0.0
Input: [0.0, 1.0] => Predicted: 0.98 | Target: 1.0
Input: [1.0, 0.0] => Predicted: 0.97 | Target: 1.0
Input: [1.0, 1.0] => Predicted: 0.03 | Target: 0.0

Training complete! 🎯

Want to really understand backpropagation and gradients?

• Dive into engine.py and explore the Value class

• Inspect how operations dynamically build a graph

• See how .backward() traverses it for gradient computation — just like real frameworks!

This project is heavily inspired by micrograd by Andrej Karpathy, licensed under the MIT License.

This project is licensed under the MIT License. See the LICENSE file for details.

Built with ❤️ by Muawiya, as part of a deep dive into AI, neural nets, and autodiff fundamentals.

• 📺 YouTube: @Coding_Moves
• 💻 GitHub: Muawiya
• 🧠 LeetCode: Moavia_Amir
• 📊 Kaggle: Moavia Amir