nanograd-aman 0.1.0

A lightweight backpropagation package for neural networks

🧠 Nanograd - Lightweight Autograd Engine for Deep Learning

Nanograd is a minimalistic automatic differentiation engine for building and training neural networks. Inspired by PyTorch’s autograd, it provides an easy-to-use framework for defining computational graphs, performing backpropagation, and training models.

🚀 Features

✅ Automatic Differentiation - Compute gradients with ease using backpropagation.
✅ Graph-Based Computation - Uses a dynamic computation graph to track operations.
✅ Lightweight & Fast - No unnecessary dependencies, optimized for speed.
✅ Custom Neural Networks - Build and train models from scratch.
✅ Graph Visualization - Visualize computational graphs using graphviz.

---

📦 Installation

You can install Nanograd directly from PyPI:

pip install nanograd

🔧 Usage

1️⃣ Defining Computation

from nanograd.engine import Value

a = Value(2.0)
b = Value(3.0)
c = a * b + 5
c.backward()

print(f"Value of c: {c.data}")        # Output: 11.0
print(f"Gradient of a: {a.grad}")     # Output: 3.0
print(f"Gradient of b: {b.grad}")     # Output: 2.0

2️⃣ Building a Neural Network

from nanograd.nn import MLP
import numpy as np

# Create a 2-layer neural network (2 inputs, 4 hidden, 1 output)
model = MLP(2, [4, 1])

# Dummy data
X = np.array([[1.0, 2.0]])
y = np.array([1.0])

# Forward pass
pred = model.forward(X)
print(f"Prediction: {pred}")

3️⃣ Visualizing Computational Graph

from nanograd.graph import draw_graph
draw_graph(c)  # Generates a graph of computations