PredictionIO-Churn-Prediction-H2O-Sparkling-Water

PredictionIO Engine integrated with Sparkling Water. Open Source project Spring 2015 @CMU.

Overview

This is an engine template for PredictionIO with Sparkling Water integration. The goal is to use Deep Learning algorithm to predict the churn rate for a phone carrier's customers.

Setup

• Please follow Install PredictionIO and Download Engine Template first for the initial setup.
• Run pio-start-all to start the PredictionIO environment.
• Run pio app new [app-name] to create a new application in PredictionIO.
• Update engine.json with the new App Name acquired from last step.
• Import the data set with python data/import_eventserver.py --access_key [app-access-key].
• Configure Deep Learning algorithm parameters in engine.json.
• Run PredictionIO (pio build && pio train && pio deploy)

Example

First, make sure that your engine is running. Then, in a Python shell, execute

 import predictionio
 engine_client = predictionio.EngineClient(url="http://localhost:8000")

in order to instantiate the engine client. After that, you can use engine_client.send_query({"param1": value1, "param2": value2, ...}) to make predictions. In order to make predictions successfully, you have to specify the following parameters:

intlPlan, voiceMailPlan, numVmailMsg, totalDayMins, totalDayCalls, totalDayCharge, totalEveMins, totalEveCalls, totalEveCharge, totalNightMins, totalNightCalls, totalNightCharge, totalIntlMins, totalIntlCalls, totalIntlCharge, customerServiceCalls

An example prediction command could be:

engine_client.send_query({'intlPlan': True, 'voiceMailPlan': True, 'numVmailMsg': 41, 'totalDayMins': 173.1, 
    'totalDayCalls': 85, 'totalDayCharge': 29.43, 'totalEveMins': 203.9, 'totalEveCalls': 107, 
    'totalEveCharge': 17.33, 'totalNightMins': 122.2, 'totalNightCalls': 78, 'totalNightCharge': 14.02, 
    'totalIntlMins': 10.0, 'totalIntlCalls': 15, 'totalIntlCharge': 3.94, 'customerServiceCalls': 0})

Note

Sparkling Water library jar is downloaded with the template (in lib folder) instead of imported from build.sbt because the maven repository is outdated.