Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/h2oai/h2o-3/llms.txt

Use this file to discover all available pages before exploring further.

H2O-3 is an open source, in-memory, distributed, fast, and scalable machine learning and predictive analytics platform. It lets you build machine learning models on big data and provides easy productionalization of those models in an enterprise environment.
H2O-3 is licensed under the Apache License, Version 2.0. Source code, issue tracking, and community discussion are available on GitHub.

​
What is H2O-3?

H2O-3 is an in-memory platform for distributed, scalable machine learning. Its core code is written in Java. A distributed key-value store is used to access and reference data, models, and objects across all nodes and machines. Algorithms are implemented on top of H2O-3’s distributed map-reduce framework and use the Java fork/join framework for multi-threading. Data is read in parallel, distributed across the cluster, stored in-memory in a columnar compressed format. H2O’s data parser has built-in intelligence to guess the schema of incoming datasets and supports data ingest from multiple sources in various formats.

​
Supported algorithms

H2O-3 includes production-ready implementations of the following algorithms:
  • AdaBoost — boosting ensemble for classification
  • AutoML — fully automatic model training and selection
  • Cox Proportional Hazards (CoxPH) — survival analysis
  • Decision Tree — single decision tree learner
  • Deep Learning — multi-layer neural networks
  • Distributed Random Forest (DRF) — tree-based ensemble
  • Distributed Uplift Random Forest — treatment effect estimation
  • Extended Isolation Forest — anomaly detection
  • Generalized Additive Models (GAM) — flexible semi-parametric models
  • Generalized Linear Model (GLM) — linear, logistic, and Poisson regression
  • Generalized Low Rank Models (GLRM) — matrix factorization
  • Gradient Boosting Machine (GBM) — tree boosting for regression and classification
  • Isolation Forest — anomaly detection via random partitioning
  • Isotonic Regression — monotone regression
  • K-Means Clustering — unsupervised partitioning
  • Naïve Bayes Classifier — probabilistic classification
  • Principal Component Analysis (PCA) — dimensionality reduction
  • RuleFit — interpretable rule-based model
  • Stacked Ensembles — meta-learner combining base models
  • Support Vector Machine (PSVM) — kernel-based classification
  • Target Encoding — categorical feature preprocessing
  • Word2Vec — word embedding from text
  • XGBoost — optimized gradient boosting

​
Multi-language support

H2O-3 exposes a consistent API across multiple languages and interfaces. All client libraries communicate with the H2O-3 backend through the REST API.

Python

Install via pip install h2o. Full-featured client with estimators, frames, and AutoML.

R

Install via install.packages("h2o"). Mirrors the Python API with idiomatic R conventions.

Flow UI

Browser-based notebook interface available at http://localhost:54321 when the cluster is running.

REST API

JSON over HTTP. All capabilities of H2O-3 are accessible from any language or tool.
Java and Scala users can access H2O-3 through the REST API or by embedding H2O-3 as a Maven artifact in their projects.

​
Architecture overview

​
Cluster model

An H2O-3 cluster is a set of JVM processes (nodes) that work together as a single distributed system. Nodes communicate peer-to-peer — there is no designated master node for data distribution.
  • Cluster formation: New H2O-3 nodes join during launch using multicast or flatfile-based discovery. Once a job starts, the cluster locks and prevents new members from joining.
  • In-memory storage: Data is stored across all nodes in a columnar compressed format. Each column (Vec) is split into contiguous subsets (Chunk) distributed across the cluster.
  • Distributed computation: MRTask (Map/Reduce) moves computation to the data rather than moving data to the computation. Results reduce up a tree back to the initiating node.

​
Distributed key-value store (DKV)

Every object — frames, models, chunks — has a home node determined by consistent hashing of its key. The DKV is used to locate and access all distributed objects: 
DKV.put(key, value)   // store an object
DKV.get(key)          // retrieve an object from its home node

​
REST API

H2O-3’s REST API allows access to all capabilities from an external program or script through JSON over HTTP. The REST API is used by:
  • The Flow web UI
  • The R binding (H2O-R)
  • The Python binding (H2O-Python)
  • Any custom integration
The default port for the REST API is 54321. The internal communication port is 54322.

​
Key sections

Quickstart

Train your first model in Python or R in under 5 minutes.

Installation

Install H2O-3 via pip, conda, CRAN, or download the standalone jar.

Algorithm reference

Detailed documentation for every supported algorithm.

AutoML

Automatically train and rank hundreds of models with a single call.

Build docs developers (and LLMs) love

Get started for freeTalk to us