Rand Stats

H2O::Client

zef:antononcube

Raku-H2O-Client

MacOS Linux Win64

Raku REST client for the open-source, distributed in-memory machine-learning platform H2O-3.

Installation

From Zef ecosystem:

zef install H20::Client

From GitHub:

zef install https://github.com/antononcube/Raku-H2O-Client.git

Setup

  1. Download the latest H20 version.

  2. Start in a OS-terminal application the H2O cluster with the shell command:

java -jar h2o.jar
  1. In case that command give the message:

Only Java versions 8-17 are supported, system version is 22.0.2

3.1. Check the available Java distributions with:

/usr/libexec/java_home -V

3.2. Pick one with a version between 8-17 or download a one with one of these versions:

3.3. Setup JAVA_HOME and run the command java -jar h2o.jar again. For example, on macOS:

export JAVA_HOME=/Library/Java/JavaVirtualMachines/zulu-11.jdk/Contents/Home 
java -jar h2o.jar

Usage examples

use H2O::Client;

my $h2o = H2O::Client.new('http://127.0.0.1:54321');

my $frame = $h2o.upload(
    [%(x => 1, group => 'a'), %(x => 2, group => 'b')],
    destination-frame => 'example.hex'
).wait.result;

say $frame.shape;       # (2 2)
say $frame.names;
say $frame<group>.type; # column proxy
say $frame.head(2);     # explicitly download a small preview

H2O::Client::Frame is a lightweight handle to a server-side frame. Its identity is the H2O frame key; metadata is fetched lazily and cached. Use refresh after external changes. H2O::Client::Column exposes column type, domain, and summary statistics. Long-running parse, model-build, and export operations return H2O::Client::Job; call wait before consuming result.

Rapids transformations

Frame transformations are explicit, composable Rapids expressions. They do not run until materialize is called with a destination key:

my $adults = $frame
    .where($frame<age>.expression.greater-than(18))
    .select(<age income>)
    .materialize('adults.hex');

The initial layer supports column selection, row filtering, arithmetic, comparisons, boolean operations, cbind, rbind, sum, and mean. It opens an H2O Rapids session lazily; call $h2o.close when retaining a connection for a long-running process is no longer necessary.

To start a local cluster:

my $h2o = H2O::Client.new;
$h2o.init(jar-path => '/path/to/h2o.jar', jvm-opts => <-Xmx4g>);
LEAVE $h2o.shutdown;

shutdown stops only a process started by this client. Shutting down a cluster connected to externally requires the explicit :cluster option.

Tests

Fast tests use mocked transport responses and do not require Java:

prove6 -Ilib t

Live integration tests are intentionally kept under xt/:

H2O_JAR=/path/to/h2o.jar prove6 -Ilib xt

H2O 3.46 supports Java 8–17. If java on PATH is newer, set H2O_JAVA to the full path of a supported Java executable.