Hierarchical Time Series Forecasting

This project provides a comprehensive pipeline for hierarchical time series forecasting. It leverages the hierarchicalforecast package to build, evaluate, and compare various forecasting models, including a novel graph-based neural network. The primary goal is to provide a robust framework for handling both temporal and geographical aggregations in time series data, with a focus on probabilistic predictions and evaluation.

Features

• End-to-End Pipeline: From data loading and preprocessing to model training, prediction, and evaluation.
• Hierarchical Forecasting: Supports both temporal and geographical (or other categorical) hierarchies.
• Probabilistic Predictions: Generates probabilistic forecasts to capture uncertainty.
• Graph-Based Neural Network: Implements a custom graph-based neural network for comparison with baseline models.
• Modular and Scalable: The codebase is structured to be easily extensible and scalable.

Project Structure

hierarchical_ts/
├── data/                 # Raw and processed data
├── models/               # Trained models
├── scripts/              # Scripts to run the pipeline
│   └── run_pipeline.py
├── src/                  # Source code
│   └── hierarchical_ts/
│       ├── __init__.py
│       ├── data.py       # Data loading and preprocessing
│       ├── models.py     # Forecasting models
│       ├── train.py      # Model training
│       ├── predict.py    # Prediction pipeline
│       ├── evaluate.py   # Evaluation metrics
│       └── utils.py      # Utility functions
├── tests/                # Unit tests
├── .gitignore
├── pyproject.toml        # Project metadata and dependencies
├── setup.cfg             # Package configuration
└── README.md

Getting Started

Prerequisites

• Python 3.8+
• Poetry (for dependency management)

Installation

1. Clone the repository:

   git clone https://your-repo-url/hierarchical_ts.git
   cd hierarchical_ts

2. Install the dependencies:

   pip install .

Usage

To run the full pipeline, execute the run_pipeline.py script:

python scripts/run_pipeline.py

This will:

1. Load and preprocess the data.
2. Train the forecasting models.
3. Generate predictions.
4. Evaluate the models and save the results.

Contributing

Contributions are welcome! Please open an issue or submit a pull request for any changes.

License

This project is licensed under the MIT License. See the LICENSE file for details.