Example: Text Analytics as a Service

This example deploys a Twitter sentiment classifier as a microservice accessible via an API POST request. The classifier was built using PL/Python in Greenplum Database (GPDB).

Read this blog post for more information.

There are two implementatations of deploying the model using

• Jupyter Notebook (via the Jupyter Kernel Gateway)
• Flask

Both apps are deployed on Cloud Foundry. This example deploys a trained sentiment classifier, but this framework can be modified for additional text analytics and general data science tasks.

Both implementations use this sentiment classifier which was built in a massively parallel processing environment using GPDB. The classifier is based on the approach of Go et al using the Sentiment140 data

Deploying to Cloud Foundry

If you do not have an account on a Cloud Foundry installation you can register for a free trial at Pivotal Web Services (PWS).

Download the Cloud Foundry Command Line Interface from the CF management console or the CF Github repo. This provides the cf command which you will use to interact with a CF installation.

Alternatively, you can install PCF Dev to install Cloud Foundry on a single work station.

Deploying the classifier

Jupyter Notebook Microservice

cd jupyter_gateway_deployment
cf push

Flask Microservice

cd flask_deployment
cf push

Using the classifier via the API

The classifier accepts POST requests of text and returns a polarity prediction from 0 to 1. Lower polarity indicates negative sentiment and higher polarity indicates positive sentiment.

$ curl -H "Content-Type: application/json" -X POST -d '{"data":["This app is awesome and in the CLOUD","Steph Curry is a basketball player","i am so mad and angry"]}' sentiment-compute-flask.cfapps.pez.pivotal.io/polarity_compute

# returned result
{
  "polarity": [
    0.7803128570201056,
    0.49108974839197184,
    0.14809403223487322
  ]
}

Examples of both apps are currently deployed and be accessed at:

sentiment-compute.cfapps.pez.pivotal.io/polarity_compute

sentiment-compute-flask.cfapps.pez.pivotal.io/polarity_compute

Resources

• Sentiment classifier
• Python Cloud Foundry examples
• Cloud Foundry Documentation
• Pivotal Blog: Model scoring as a service

Author

Chris Rawles