h2o-scoring-engine-base

The h2o-scoring-engine-base is an application prototype for exposing H2O model POJO as a RESTful service. Repository contains a prototype application and a building script for creating scoring engine based on provided H2O model.

Prerequisites

You'll need a JDK version 1.8, and also to have the parent pom installed with the correct version:

git clone https://github.com/trustedanalytics/trustedanalytics-cloud-parent.git
cd trustedanalytics-cloud-parent
git checkout edf7681e
mvn clean install

You'll also need to ensure that your model is named such that it is a valid Java class name. For example, use underscores (_) rather than hyphens (-).

Building scoring engine

To build h2o-scoring-engine-base, run the following command from the project root directory:

mvn clean package

After building h2o-scoring-engine-base go to tools directory and run a h2o-scoring-engine-builder.sh passing 3 parameters: H2O server URL, H2O model name and a path to previously built h2o-scoring-engine-base JAR.

 ./h2o-scoring-engine-builder.sh https://<h2o server> <model_name> ../target/h2o-scoring-engine-base.jar

Optionally, you can also pass output file name and credentials for H2O server.

./h2o-scoring-engine-builder.sh  -o <output jar name> -u <username> -p https://<h2o server> <model name> ../target/h2o-scoring-engine-base.jar

Builder script downloads your model and a required h2o-genmodel.jar library form H2O server and builds a scoring engine based on the model. Application JAR, h2o-scoring-engine-<model name>.jar, is placed in working directory.

To run it locally run:

java -jar h2o-scoring-engine-<model name>.jar

To specify a particular port, rather than the default "8080," append to the above command --server.port <port>.

Using scoring engine

Scoring engine score method is accessible through REST API:

URL: http://<application host>/score

Headers: Content-type: application/json

HTTP Method: POST

Request body: sequence of numbers of a size required by the model

Usage example:

curl -i -X POST -H "Content-type: application/json" -d '[1.2, 1.3, 1.4, 5.3]' http://<application host>/score

URL: http://<application host>/score-object

Headers: Content-type: application/json

HTTP Method: POST

Request body: JSON object of name:value pairs.

Usage example:

curl -i -X POST -H "Content-type: application/json" -d '{"foo": "bar", "baz": "1"}' http://<application host>/score-object

URL: http://<application host>/names

HTTP Method: GET

Response: JSON Array of names supported by the model for scoring with a JSON Object.