PredictionIO Iris Template

Requirement

• Spark 2.x

Getting Started

Create App on PIO

Create IrisApp application on PredictionIO:

pio app new --access-key IRIS_TOKEN IrisApp

Import Data

Import iris data from iris.data.

python data/import_eventserver.py

Build Template

Build this template:

pio build

Run Jupyter

Launch Jupyter notebook and open eda.ipynb. (or you can create a new notebook to analyze data)

PYSPARK_PYTHON=$PYENV_ROOT/shims/python PYSPARK_DRIVER_PYTHON=$PYENV_ROOT/shims/jupyter PYSPARK_DRIVER_PYTHON_OPTS="notebook" pio-shell --with-pyspark

Run Train on Spark

Download Python code from eda.ipynb and put it to train.py. To execute it on Spark, run pio train with --main-py-file option.

pio train --main-py-file train.py

Deploy App

Run PredictionIO API server:

pio deploy

Predict

Check predictions from deployed model:

curl -s -H "Content-Type: application/json" -d '{"attr0":5.1,"attr1":3.5,"attr2":1.4,"attr3":0.2}' http://localhost:8000/queries.json

Run on Docker

PredictionIO Setup dockernizes PredictionIO Template project easily and provides All-In-One environment for train and deploy processes.

Run Train Steps on Docker

train mode executes create-app, import-data, build and train process.

PIO_MODE=train docker-compose up --abort-on-container-exit

Run Deploy Step on Docker

deploy mode starts Rest API on Docker container.

PIO_MODE=deploy docker-compose up --abort-on-container-exit

and then check a prediction:

curl -s -H "Content-Type: application/json" -d '{"attr0":5.1,"attr1":3.5,"attr2":1.4,"attr3":0.2}' http://localhost:8000/queries.json