This repository implements a custom artificial neural network from scratch using PyTorch. The network architecture is built without relying on pre-defined layers like nn.Linear or activation functions like nn.Tanh. It is trained on a regression dataset and achieves an accuracy of 87%, outperforming a Keras model trained on the same data (84% accuracy).
Builds the neural network architecture from scratch using PyTorch tensors and operations.
PyTorch for automatic differentiation and efficient loss calculation.
Designed for regression tasks, demonstrating strong performance on a chosen regression dataset.
Provides a hands-on exploration of neural network fundamentals by building the components from scratch.
Offers the flexibility to tailor the network architecture and activation functions to specific problem requirements.
Achieves competitive results compared to models built with higher-level abstractions.
Bash
git clone https://github.com/<your-username>/from-scratch-nn-pytorch.git
Bash
pip install torch
# Additional dependencies for your specific dataset (if required)
Bash
python train.py
This script loads the dataset, trains the model, and evaluates its performance.
Explore different network architectures and activation functions.
Implement additional features like early stopping and regularization.
Test the model on various regression datasets.
We welcome pull requests and suggestions for improvement.