BentoProphet

A time series forecasting project demonstrating how to build inference APIs using Facebook's Prophet model with BentoML. This project provides two different approaches to serving Prophet models for time series forecasting.

📋 Overview

BentoProphet showcases two different patterns for serving time series forecasting models:

• fit_predict/: Train-and-predict service that trains a new Prophet model on each request
• predict/: Pre-trained model service that loads a saved model and makes predictions

Both services provide APIs for:

• Generating forecasts with confidence intervals
• Creating forecast visualization plots
• Analyzing forecast components (trend, seasonality, etc.)

🔧 Getting Started

Option 1: Fit-and-Predict Service

Navigate to the fit_predict/ directory for a service that trains models dynamically on each request.

Option 2: Pre-trained Model Service

Navigate to the predict/ directory for a service that uses pre-trained models for faster inference.

🌐 Deployment

Both services can be deployed locally using bentoml serve or to BentoCloud for production use.

📖 Documentation

For detailed usage instructions and examples, see the README files in each service directory:

• Fit-Predict Service Documentation
• Pre-trained Model Service Documentation

🤝 Contributing

This project is part of the BentoML examples collection. For contributions and issues, please refer to the BentoML documentation.

📄 License

This project is licensed under the MIT License.