Apache PredictionIO Docker

Overview

PredictionIO Docker provides Docker image for use in development and production environment.

Usage

Run PredictionIO with Selectable docker-compose Files

You can choose storages for event/meta/model to select docker-compose.yml.

docker-compose -f docker-compose.yml -f ... up

Supported storages are as below:

If you run PredictionIO with Postgresql, run as below:

docker-compose -f docker-compose.yml \
  -f pgsql/docker-compose.base.yml \
  -f pgsql/docker-compose.meta.yml \
  -f pgsql/docker-compose.event.yml \
  -f pgsql/docker-compose.model.yml \
  up

To use localfs as model storage, change as below:

docker-compose -f docker-compose.yml \
  -f pgsql/docker-compose.base.yml \
  -f pgsql/docker-compose.meta.yml \
  -f pgsql/docker-compose.event.yml \
  -f localfs/docker-compose.model.yml \
  up

Tutorial

In this demo, we will show you how to build a recommendation template.

Run PredictionIO environment

The following command starts PredictionIO with an event server. PredictionIO docker image mounts ./templates directory to /templates.

$ docker-compose -f docker-compose.yml \
    -f pgsql/docker-compose.base.yml \
    -f pgsql/docker-compose.meta.yml \
    -f pgsql/docker-compose.event.yml \
    -f pgsql/docker-compose.model.yml \
    up

We provide pio-docker command as an utility for pio command. pio-docker invokes pio command in PredictionIO container.

$ export PATH=`pwd`/bin:$PATH
$ pio-docker status
...
[INFO] [Management$] Your system is all ready to go.

Download Recommendation Template

This demo uses predictionio-template-recommender.

$ cd templates
$ git clone https://github.com/apache/predictionio-template-recommender.git MyRecommendation
$ cd MyRecommendation

Register Application

You need to register this application to PredictionIO:

$ pio-docker app new MyApp1
[INFO] [App$] Initialized Event Store for this app ID: 1.
[INFO] [Pio$] Created a new app:
[INFO] [Pio$]       Name: MyApp1
[INFO] [Pio$]         ID: 1
[INFO] [Pio$] Access Key: i-zc4EleEM577EJhx3CzQhZZ0NnjBKKdSbp3MiR5JDb2zdTKKzH9nF6KLqjlMnvl

Since an access key is required in subsequent steps, set it to ACCESS_KEY.

$ ACCESS_KEY=i-zc4EleEM577EJhx3CzQhZZ0NnjBKKdSbp3MiR5JDb2zdTKKzH9nF6KLqjlMnvl

engine.json contains an application name, so replace INVALID_APP_NAME with MyApp1.

...
"datasource": {
  "params" : {
    "appName": "MyApp1"
  }
},
...

Import Data

To import training data to Event server for PredictionIO, this template provides an import tool. The tool depends on PredictionIO Python SDK and install as below:

$ pip install predictionio

and then import data:

$ curl https://raw.githubusercontent.com/apache/spark/master/data/mllib/sample_movielens_data.txt --create-dirs -o data/sample_movielens_data.txt
$ python data/import_eventserver.py --access_key $ACCESS_KEY

Build Template

This is Scala based template. So, you need to build this template by pio command.

$ pio-docker build --verbose

Train and Create Model

To train a recommendation model, run train sub-command:

$ pio-docker train

Deploy Model

If a recommendation model is created successfully, deploy it to Prediction server for PredictionIO.

$ pio-docker deploy

You can check predictions as below:

$ curl -H "Content-Type: application/json" \
-d '{ "user": "1", "num": 4 }' http://localhost:8000/queries.json

Advanced Topics

Run with Elasticsearch

For Elasticsearch, Meta and Event storage are available. To start PredictionIO with Elasticsearch,

docker-compose -f docker-compose.yml \
  -f elasticsearch/docker-compose.base.yml \
  -f elasticsearch/docker-compose.meta.yml \
  -f elasticsearch/docker-compose.event.yml \
  -f localfs/docker-compose.model.yml \
  up

Run with Spark Cluster

Adding docker-compose.spark.yml, you can use Spark cluster on pio train.

docker-compose -f docker-compose.yml \
  -f docker-compose.spark.yml \
  -f elasticsearch/docker-compose.base.yml \
  -f elasticsearch/docker-compose.meta.yml \
  -f elasticsearch/docker-compose.event.yml \
  -f localfs/docker-compose.model.yml \
  up

To submit a training task to Spark Cluster, run pio-deploy train with --master option:

pio-docker train -- --master spark://spark-master:7077

See docker-compose.spark.yml if changing settings for Spark Cluster.

Run Engine Server

To deploy your engine and start an engine server, run Docker with docker-compose.deploy.yml.

docker-compose -f docker-compose.yml \
  -f pgsql/docker-compose.base.yml \
  -f pgsql/docker-compose.meta.yml \
  -f pgsql/docker-compose.event.yml \
  -f pgsql/docker-compose.model.yml \
  -f docker-compose.deploy.yml \
  up

See deploy/run.sh and docker-compose.deploy.yml if changing a deployment.

Run with Jupyter

You can launch PredictionIO with Jupyter.

docker-compose -f docker-compose.jupyter.yml \
  -f pgsql/docker-compose.base.yml \
  -f pgsql/docker-compose.meta.yml \
  -f pgsql/docker-compose.event.yml \
  -f pgsql/docker-compose.model.yml \
  up

For more information, see JUPYTER.md.

Development

Build Base Docker Image

docker build -t predictionio/pio pio

Build Jupyter Docker Image

docker build -t predictionio/pio-jupyter jupyter

Push Docker Image

docker push predictionio/pio:latest
docker tag predictionio/pio:latest predictionio/pio:$PIO_VERSION
docker push predictionio/pio:$PIO_VERSION