by chanlee514


PredictionIO classification app

Predictive classification powered by PredictionIO, machine learning on Heroku.

This is a demo application of PredictionIO preset for simplified deployment. Custom PredictionIO engines may be deployed as well, see CUSTOM documentation.

Once deployed, this engine demonstrates prediction of the best fitting service plan for a mobile phone user based on their voice, data, and text usage. The model is trained with a small, example data set.

How To 📚

✏️ Throughout this document, code terms that start with $ represent a value (shell variable) that should be replaced with a customized value, e.g $eventserver_name, $engine_name, $postgres_addon_id

Deploy to Heroku

Please follow steps in order.

  1. Requirements
  2. Eventserver
    1. Create the eventserver
    2. Deploy the eventserver
  3. Classification engine
    1. Create the engine
    2. Connect the engine with the eventserver
    3. Import data
    4. Deploy the engine


Once deployed, how to work with the engine.

Deploy to Heroku 🚀

1. Requirements

2. Eventserver

Create the eventserver

git clone \ \

cd pio-eventserver

heroku create $eventserver_name
heroku addons:create heroku-postgresql:hobby-dev
heroku buildpacks:add -i 1
heroku buildpacks:add -i 2 heroku/scala

Deploy the eventserver

We delay deployment until the database is ready.

heroku pg:wait && git push heroku master

3. Classification Engine

We'll be using a classification engine for Heroku which implements Spark's Random Forests algorithm to predict a label using decision trees. See A Visual Introduction to Machine Learning to learn why decision trees are so effective.

(Originally this engine implemented Spark's Naive Bayes algorithm. We soon switched to Random Forests to improved results by correlating attributes, a well-known weakness of Naive Bayes. The Bayes algorithm is still available in the engine source.)

Create the engine

git clone \ \

cd pio-engine-classi

heroku create $engine_name
heroku buildpacks:add -i 1
heroku buildpacks:add -i 2

Connect the engine with the eventserver

First, collect a few configuration values.

Get the eventserver's database add-on ID

heroku addons:info heroku-postgresql --app $eventserver_name
# Use the returned Postgres add-on ID
# to attach it to the engine.
# Example: `postgresql-aerodynamic-00000`
heroku addons:attach $postgres_addon_id --app $engine_name

Get an access key for this engine's data

heroku run 'pio app new classi' --app $eventserver_name
# Use the returned access key for `$pio_app_access_key`
heroku config:set \
  PIO_EVENTSERVER_ACCESS_KEY=$pio_app_access_key \

Import data

🚨 Mandatory: data is required for training. The model cannot answer predictive queries until trained with data.

When deployed, the engine will automatically train a model to predict the best fitting service plan for a mobile phone user based on their voice, data, and text usage. We'll use the engine's example data and import script for initial training.

  • pip install predictionio may be required before the import script will run; see how-to install pip
python ./data/ \
  --url https://$ \
  --access_key $pio_app_access_key

Deploy the engine

git push heroku master

# Follow the logs to see training 
# and then start-up of the engine.
heroku logs -t --app $engine_name

Usage ⌨️

Scale up

Once deployed, scale up the processes and config Spark to avoid memory issues. These are paid, professional dyno types:

heroku ps:scale \
  web=1:Standard-2X \
  release=0:Performance-L \
  train=0:Performance-L \
  --app $engine_name

Query for predictions

Once deployment completes, the engine is ready to predict the best fitting service plan for a mobile phone user based on their voice, data, and text usage.

Submit queries containing these three user attributes to get predictions using Spark's Random Forests algorithm:

# Fits low usage, `0`
curl -X "POST" "https://$" \
     -H "Content-Type: application/json; charset=utf-8" \
     -d "{\"voice_usage\":12,\"data_usage\":0,\"text_usage\":4}"

# Fits more voice, `1`
curl -X "POST" "https://$" \
     -H "Content-Type: application/json; charset=utf-8" \
     -d "{\"voice_usage\":480,\"data_usage\":0,\"text_usage\":121}"

# Fits more data, `2`
curl -X "POST" "https://$" \
     -H "Content-Type: application/json; charset=utf-8" \
     -d "{\"voice_usage\":25,\"data_usage\":1000,\"text_usage\":80}"

#Fits more texts, `3`
curl -X "POST" "https://$" \
     -H "Content-Type: application/json; charset=utf-8" \
     -d "{\"voice_usage\":5,\"data_usage\":80,\"text_usage\":1000}"

#Extreme voice & data, `4`
curl -X "POST" "https://$" \
     -H "Content-Type: application/json; charset=utf-8" \
     -d "{\"voice_usage\":450,\"data_usage\":1104,\"text_usage\":43}"

#Extreme data & text, `5`
curl -X "POST" "https://$" \
     -H "Content-Type: application/json; charset=utf-8" \
     -d "{\"voice_usage\":24,\"data_usage\":770,\"text_usage\":482}"

#Extreme voice & text, `6`
curl -X "POST" "https://$" \
     -H "Content-Type: application/json; charset=utf-8" \
     -d "{\"voice_usage\":450,\"data_usage\":80,\"text_usage\":332}"

#Everything equal / balanced usage, `7`
curl -X "POST" "https://$" \
     -H "Content-Type: application/json; charset=utf-8" \
     -d "{\"voice_usage\":450,\"data_usage\":432,\"text_usage\":390}"

For a production model, more aspects of a user account and their correlations might be taken into consideration, including: account type (individual, business, or family), frequency of roaming, international usage, device type (smart phone or feature phone), age of device, etc.


If you hit any snags with the engine serving queries, check the logs:

heroku logs -t --app $engine_name

If errors are occuring, sometimes a restart will help:

heroku restart --app $engine_name

Going Deeper 🔬

This is a sample application of PredictionIO, preset to get up-and-running quickly.

Custom PredictionIO engines may be deployed with this buildpack too. See CUSTOM documentation including: training, evaluation, & configuration.