A work-in-progress machine learning project (PyTorch)
This project explores the use of Long Short-Term Memory (LSTM) neural networks to model and forecast precipitation based on historical time-series data.
The goal is to evaluate how well recurrent architectures can capture temporal patterns in rainfall and improve short-term prediction accuracy.
Although the project is still under development, the repository already includes the initial data processing pipeline, baseline models, and early LSTM prototypes.
- Build a reproducible precipitation time-series dataset
- Train and evaluate a baseline LSTM model
- Experiment with different architectures (stacked LSTM, bidirectional LSTM, etc.)
- Compare performance with simpler baselines (persistence, linear regression, etc.)
- Visualize predictions vs. ground truth
The project is in progress and updated during my free time.
Model experiments, data preparation routines, and evaluation metrics will be expanded over time.
- Python >3.9
- PyTorch
- NumPy
- Pandas
- Matplotlib
(Additional dependencies will be added as the project evolves.)