Skip to main content
H2O-3 runs on Java. You must have a supported JRE installed before using the Python or R packages, the standalone jar, or any other distribution. The client libraries (h2o for Python, h2o for R) communicate with the H2O-3 backend over HTTP and do not embed Java themselves.

​
Requirements

ComponentMinimum version
Java (JRE)Java SE 8 (64-bit)
Python3.7, 3.8, 3.9, 3.10, or 3.11
R3.1 or later
Operating systemWindows 7+, macOS 10.9+, Ubuntu 12.04+, RHEL/CentOS 6+
H2O-3 requires 64-bit Java. The 64-bit JDK is required to build H2O-3 or run tests. The 64-bit JRE is sufficient to run the H2O-3 binary or the Python/R packages. H2O-3 supports Java SE 8 through 17.
H2O-3 also requires numpy<2. If you have numpy 2 installed, run: 
pip install --force-reinstall 'numpy<2'

​
Install in Python

1

Install via pip

The simplest installation method is pip. This installs the latest stable release.
pip install h2o
To upgrade an existing installation:
pip install --upgrade h2o
2

Install via conda (Anaconda)

Install H2O-3 from the h2oai conda channel. Leave the version blank to get the latest release.
conda install -c h2oai h2o
For Python 3.6+ users, the tabulate dependency is available through the conda-forge channel. Add it before installing:
conda create -n h2o_env python=3.9 anaconda
source activate h2o_env
conda config --append channels conda-forge
conda install -c h2oai h2o
3

Install a specific stable version

To install directly from the H2O release server (guarantees the latest stable version, which may lag behind on PyPI):
pip install -f http://h2o-release.s3.amazonaws.com/h2o/latest_stable_Py.html h2o
4

Verify the installation

Start Python and initialize H2O-3. A successful connection prints the cluster status table.
import h2o
h2o.init()
Expected output includes lines similar to:
H2O cluster version:        3.47.0
H2O cluster total nodes:    1
H2O cluster free memory:    X Gb
H2O cluster total cores:    N
H2O cluster status:         accepting new members, healthy
H2O connection url:         http://127.0.0.1:54321

​
Install in R

1

Remove any existing H2O-3 installation

If you have a previous version installed, remove it first to avoid conflicts.
if ("package:h2o" %in% search()) { detach("package:h2o", unload = TRUE) }
if ("h2o" %in% rownames(installed.packages())) { remove.packages("h2o") }
2

Install dependencies

Install the packages that H2O-3 depends on.
pkgs <- c("RCurl", "jsonlite")
for (pkg in pkgs) {
  if (!(pkg %in% rownames(installed.packages()))) { install.packages(pkg) }
}
3

Install from CRAN

Install the latest version of H2O-3 from CRAN.
install.packages("h2o")
The CRAN release may lag slightly behind the latest stable version. To guarantee the latest stable release, install directly from the H2O release server instead:
install.packages(
  "h2o",
  type   = "source",
  repos  = c("http://h2o-release.s3.amazonaws.com/h2o/latest_stable_R")
)
4

Verify the installation

Load the package and initialize H2O-3.
library(h2o)
h2o.init()
Successful output includes:
R is connected to the H2O cluster:
  H2O cluster version:        3.47.0
  H2O cluster total nodes:    1
  H2O cluster total memory:   X.XX GB
  H2O cluster healthy:        TRUE
  H2O connection ip:          localhost
  H2O connection port:        54321

​
Download the standalone H2O jar

Use the standalone jar when you want to run H2O-3 from the command line, start a multi-node cluster, or connect to it from any language via the REST API.
1

Download the jar

Download the latest stable release zip from the H2O download page:
# Download the zip (replace with the actual current version from h2o.ai/download)
curl -O https://h2o-release.s3.amazonaws.com/h2o/latest_stable.html
Visit https://h2o.ai/download to find the direct download link for the current stable release.
2

Unzip and start H2O-3

Unzip the archive and launch the server. By default, H2O-3 starts on port 54321.
unzip h2o-*.zip
cd h2o-*/
java -jar h2o.jar
Open http://localhost:54321 in a browser to access the Flow web UI.
3

Configure memory and cluster name

Use JVM flags to configure heap size. The rule of thumb is to allocate at least 4× the size of your dataset.
# Start with 4 GB heap
java -Xmx4g -jar h2o.jar

# Start with 6 GB heap and a named cluster
java -Xmx6g -jar h2o.jar -name MyCluster
To form a multi-node cluster, start several JVM processes with the same cluster name on reachable hosts. H2O-3 uses multicast discovery by default. For cloud environments, use a flatfile listing IP:PORT pairs and pass it with -flatfile <file>.

​
Install on Hadoop

H2O-3 can run as a YARN application on Hadoop clusters. Pre-built Hadoop zip files are available for major distributions including Cloudera CDH 5.4+, MapR 4.0+, and IBM Open Platform 4.2.
When running H2O-3 on Hadoop, use Java 8 or Java 11. Hadoop itself only supports those versions.
# Download the Hadoop zip for your distribution from https://h2o.ai/download
unzip h2o-*-hdp2.6.zip
cd h2o-*-hdp2.6/

# Launch a 1-node cluster with 6 GB per mapper
hadoop jar h2odriver.jar -nodes 1 -mapperXmx 6g
After the cluster forms, the output shows the URL to open H2O Flow in your browser: 
H2O cluster (1 nodes) is up
Open H2O Flow in your web browser: http://172.16.2.188:54321

​
Install on Spark (Sparkling Water)

Sparkling Water combines Apache Spark with H2O-3, making H2O-3’s algorithms accessible from Spark workflows. Pre-built packages are available on the download page. Resources:

​
Verifying the installation

After installing via any method, verify H2O-3 is working correctly: 
import h2o
h2o.init()

# Import a sample dataset and confirm parsing works
df = h2o.import_file(
    "https://s3.amazonaws.com/h2o-public-test-data/smalldata/iris/iris_wheader.csv"
)
print(df.shape)   # Should print (150, 5)
print(df.head())
If h2o.init() completes without error and df.shape (Python) or dim(df) (R) returns (150, 5), your installation is working correctly.

​
Next steps

Quickstart

Train your first model end-to-end in Python or R.

Introduction

Learn about H2O-3’s architecture, algorithms, and multi-language API.

Build docs developers (and LLMs) love

Get started for freeTalk to us