KittenCN/stock_prediction

What it solves

This project provides a framework for experimenting with stock price time-series prediction. It handles the entire pipeline from data acquisition and feature processing to model training, inference, and evaluation, allowing researchers to test various deep learning architectures on financial data.

How it works

The system uses PyTorch to implement a variety of prediction models. It fetches market data via APIs (such as akshare), preprocesses the data into aplayable format, and trains models to predict future price movements. To ensure accuracy during inference, the project saves normalization parameters alongside model weights to prevent feature dimension mismatches.

Who it’s for

It is designed for researchers and developers interested in financial time-series forecasting and deep learning experiments.

Highlights

• Diverse Model Support: Supports a wide range of architectures including LSTM, Transformer, Hybrid, PTFT/VSSM, Diffusion, and Graph models.
• Complete Pipeline: Includes integrated scripts for data scraping, preprocessing, training, and prediction.
• Robust Validation: Features a comprehensive testing suite with unit and integration tests, and CI integration via GitHub Actions.
• Flexible Configuration: Uses a YAML-based configuration system for managing features and file paths.