README.md

H2O AI Restful Wrapper

The purpose of this project is to provide a re-usable Restful wrappper for the H2O AI generated models.

Setup

This library requires three dependencies that will come out of your model generation:

1. Working Maven Installation
2. The h2o-genmodel.jar
3. The model deployment package (MOJO).

Loading h2o-genmodel.jar

Download the file from H2O AI.

Copy the file into $PROJECT_HOME/lib/

Download the Model Deployment Package (MOJO)

Download the file from H2O AI.

Build

Once the h2o-genmodel.jar is in place you can execute the build.

maven clean package -Dmodel-name=ransley-test-model

where ransley-test-model is some meaningful description of your model.

This will produce a JAR file in the target directory - target/ransley-test-model.jar.

Execution

This is a two step process that starts the application:

java -jar target/ransley-test-model.jar -model.path=mojo.zip

NOTE: The jar file itself can be copied around and does not have to live in the target directory. Additionally, mojo.zip is the zip file downloaded above in the Model Deployment Package (MOJO) download step.

Execution of this will look as follows:

Once this is running you can post to https://localhost:8080. The payload is expected to be JSON with name value pairs and each value is assumed to be a Double. Additionally, at the moment this only supports regressions, but I might extend this in the future.

An example of a curl (obviously the input features will be dependent on your model) is as follows:

curl -0 -v -X POST https://localhost:8080/regression \
-H "Accept: application/json" -H "Content-type: application/json" \
-X POST -d @- << EOF
'{
"past_pct_eurchf_close": -8.007584194686945, 
"past_pct_eurgbp_close": -20.86890533105512, 
"past_pct_eurjpy_close": 17.419864785013228, 
"past_pct_eurusd_close": -16.47598097 , 
"past_pct_usdcad_close": -27.46739146, 
"past_pct_usdchf_close": 8.899911001, 
"past_pct_usdjpy_close": 34.01672489, 
"past_pct_usdaud_close": -71.72595738, 
"past_pct_usdgbp_close": -4.820978727, 
"past_pct_usdnzd_close": -54.51713396
}'
EOF

The result (assuming it isn't an error) will be a JSON response with the prediction together with the inputs:

{
    input": {
        "past_pct_usdjpy_close":34.01672489,
        "past_pct_usdaud_close":-71.72595738,
        "past_pct_eurjpy_close":17.419864785013228,
        "past_pct_eurchf_close":-8.007584194686945,
        "past_pct_usdcad_close":-27.46739146,
        "past_pct_usdgbp_close":-4.820978727,
        "past_pct_eurgbp_close":-20.86890533105512,
        "past_pct_eurusd_close":-16.47598097,
        "past_pct_usdnzd_close":-54.51713396,
        "past_pct_usdchf_close":8.899911001
    },
    "prediction":1.2538692149339385
}