H2O integration

H2O is one of the leading machine learning platforms. It features an open-source server with in-memory implementations of multiple popular algorithms. Users can access the server through hosted visual notebooks or using multiple programming languages, including R and Python. AutoML functionality helps to automatically select matching model types for provided datasets. Created models can be deployed as POJOs or MOJOs (Model Object, Optimized - an alternative storage standard by H2O authors) for scoring. H2O provides both a server endpoint and a set of Java libraries for automated scoring based on deployment files.

H2O in Prinz

Prinz uses deployment files in MOJO format. Client shipped with Prinz allows for loading files from HTTP or local path.

Sample environment presents a minimal deployment:

• Models .zip files are stored on the server (see /dev-environment/h2o/exports/, empty before training);
• Server exposes list of files as a website. Minimal Python implementation lists files as a simple website at localhost:5200;
• Application config includes H2O repository;
• Nussknacker configuration file sets endpoint URL;