scottcode / moves

Data Science Demo: Real-time model scoring as a service using Pivotal Cloud Foundry

Pivotal Real-Time Data Science: Scoring-as-a-Service

moves

This application demonstrates real-time model scoring as a service using Pivotal Cloud Foundry (PCF), Pivotal Big Data Suite, Spring Cloud Data Flow, and Python-based open source machine learning. The pipeline applies broadly and would allow us to evaluate and score almost any feed of streaming data - from sensor data to unstructured text data - to drive real-time action.

Take a look at this blog post and the about page for more information.

Pre-requisites

• Pivotal Cloud Foundry
  • Redis service

Deploying the app on Pivotal Cloud Foundry

1. Update the 3 application names in manifest.yml

These app names will become part of the domain URLs, so change as desired.

...
name: DASHBOARD-APP-NAME
...
name: TRAINING-APP-NAME
...
name: SCORING-APP-NAME
...

Note that underscores are not allowed in the app names. Cloud Foundry automatically converts them to dashes, which disrupts URL routing.

2. Update parameters in JavaScript

Edit file "moves-app/moves/static/js/movesParams.js" to reflect route names of training and scoring applications as specified in previous step.

3. Create redis service and push application

cf create-service p.redis cache-small moves-redis
cf push

This has been tested using Pivotal Web Services PWS

http://docs.run.pivotal.io/devguide/deploy-apps/deploy-app.html

Contact

Chris Rawles is the original author.

For more information, please contact Scott Hajek ([e-mail]) and Jarrod Vawdrey ([e-mail])