Overview

Sparkling Water integrates H2O’s in-memory, high-performance machine learning engine with Apache Spark’s powerful data processing capabilities. It enables seamless interoperability: developers can query data with Spark SQL, build and deploy models with H2O algorithms, and run scoring pipelines all within Spark. You get Spark’s flexible APIs together with H2O’s suite of ML algorithms—scalable, fast, and easy to deploy.

Features and Capabilities

• Integrated ML & Data Engine: Combines Apache Spark’s data processing with H2O’s in-memory ML algorithms for distributed computing.
• Access to H2O Algorithms: Scalable implementations of Random Forest, GLM, GBM, XGBoost, GLRM, Word2Vec, and more.
• Language Support: Develop in Scala, R, or Python; integrate H2O Flow UI.
• Seamless API Integration: Transparent API to interchange between Spark DataFrames and H2O Frames.
• Deployment Options: Export models as optimized POJOs/MOJOs to deploy in any environment, including Spark pipelines.
• Dual Backend Modes: Internal mode launches H2O within Spark executors; external mode connects Spark to standalone H2O clusters.
• Advanced ML Workflow Support: Supports grid search, hyperparameter tuning, model ensembling (stacked learners), and provides real‑time training visualizations.
• Enterprise Features: Premade packages for major Spark versions; works with cloud services like Databricks, AWS & Azure.
• Community & Cloud Support: Backed by H2O.ai community, with enterprise support options available.