This repository provides a collection of custom Python recipes for H2O Driverless AI, enabling users to extend its automated machine learning capabilities. It caters to domain scientists and ML engineers seeking greater control over feature engineering, model building, and scoring within the Driverless AI platform. The recipes offer custom transformers, models, and scorers that can be dynamically uploaded and integrated without restarting the Driverless AI instance.

How It Works

The core concept is "Bring Your Own Recipe" (BYOR), allowing users to inject custom Python code as plugins. These recipes can implement new data transformations, model algorithms, or evaluation metrics. Driverless AI seamlessly integrates these custom components into its automated pipeline, offering an alternative or supplement to its built-in functionalities. This approach enhances precision and allows for domain-specific optimizations.

Quick Start & Requirements

• Recipes are uploaded directly into the Driverless AI interface.
• Requires a running instance of H2O Driverless AI.
• Recipes themselves are Python scripts, often leveraging libraries like pandas, numpy, scikit-learn, and H2O-3.

Highlighted Details

• Extensive library of over 200 recipes covering Data, NLP, Unsupervised learning, Explainers, Time Series, and custom Scorer implementations.
• Includes recipes for integrating with external services (e.g., BigQuery, ElasticSearch, Azure Speech) and specialized data types (e.g., audio, images, IP addresses, PE files).
• Features advanced MLI (Machine Learning Interpretability) explainers and custom loss functions for models like LightGBM and XGBoost.
• Provides templates for creating new custom transformers, models, and scorers.

Maintenance & Community

• The repository is maintained by H2O.ai.
• Links to the Driverless AI community Slack channel are provided for BYOR-related questions.

Licensing & Compatibility

• The repository itself appears to be under a permissive license (likely MIT, based on common H2O.ai open-source practices, though not explicitly stated in the README).
• Recipes are designed for use within H2O Driverless AI, which is an enterprise product. Compatibility is tied to Driverless AI versions.

Limitations & Caveats

• Recipes execute with the same privileges as the Driverless AI application, necessitating careful security review.
• The README strongly emphasizes security considerations, including potential access to all data and system resources, and recommends enabling security analysis features within Driverless AI.