Using only PyTorch's matrix multiplications, and no other built-in APIs to understand the math behind!
A self project to learn in depth about machine learning, its pitfalls, why the architectures are designed the way they are, and how to improve them.
This project is an extension for the NewsSwipe app I built in January 2024. (checkout the app here). The main goal of this project is to categorize news articles into different categories. (e.g. sports, politics, technology, etc.)
I'm using different machine learning techniques and deep learning to achieve this goal.
I've only implemented a simple RNN model and a LSTM model from scratch (using only pytorch) till now, and achieving accuracy of 83% and 86% respectively on the test dataset.
Planning to implement more complex models and techniques in the future while investigating the pitfalls of each model and documenting them.
The dataset used for this project is the "AG News" dataset. It is a collection of news articles from the AG's corpus of news articles on the web. The dataset contains 120,000 training samples and 7,600 testing samples from 42 different classes.
The dataset can be found here, in which orignal authors implemented a ConvNet on character-level inputs. The paper can be found here.
- Python 3.6+
- Pytorch
- Numpy