REST API -- Iris Flower Species Classification Model

Example repo for deploying a machine learning model as a REST API

This is our first example of being able to take a static model and deploy it to have a live inference REST API using Datmo. The goal of this is to show how easy it is to deploy a machine learning model with Datmo in a platform agnostic fashion -- deploy on AWS, Azure, GCP, or even baremetal servers, with no need to struggle over compatability issues beyond the REST spec and your package installs in the Dockerfile.

This is a simple model that uses scikit-learn in python3 to predict species for a given flower from the [Fisher Iris Flower Dataset].

Getting Started

To understand how to adapt this code to your Datmo model, refer to the blog post [here] not yet available.

Try it out (*Sorry, this endpoint is no longer live!)

Make a prediction using data

curl ec2-54-183-245-15.us-west-1.compute.amazonaws.com:8000/predicts -L -X POST -d '{"sepal_length": [6.9], "sepal_width": [3.2], "petal_length": [5.7], "petal_width": [2.3]}' -H 'Content-type: application/json'

Prediction Response: "{'Species': 'Iris-virginicia'}"

Deployment

• SSH into your server and install Datmo.
• Restart your terminal (this allows Docker to have sudo permissions)
• Use Datmo to clone the model with:

$ datmo clone nmwalsh/datmo_falcon_api

• $ screen
• $ cd datmo_falcon_api
• Initialize the gunicorn server with:

$ datmo task run "/usr/local/bin/gunicorn --access-logfile - -b 0.0.0.0:8000 falcon_gateway:app" --port 8000

—access-logfile is for enabling standard output for connection logging in gunicorn, which is disabled by default

-b 0.0.0.0:8000 binds gunicorn to port 8000 of the 0.0.0.0 localhost on the container that the model and gunicorn is running. This is designating which port gunicorn should be looking at to receive a data stream from inside of the container.

falcon_gateway:app allows gunicorn to bind our falcon python app to be recognized as the official WSGI app to serve requests and responses to/from the server.

—port 8000 is run to open up port 8000 on the docker container, so that it can receive requests from the physical server and return them as well. This is the outward facing port of the container.

API Endpoints:

The model is currently a live API with the following resources:

GET  | /info
GET  | /predicts
POST | /predicts -d {"key":[val]}

Note: This basic API doesn't have robust error handling, and will return a generic internal server error if you attempt to run a request with...

• unaccepted datatypes
• incomplete number of inputs (requires all 4 inputs)
• invalid JSON input structure.

Valid Requests:

View general info about the API

curl ec2-54-183-245-15.us-west-1.compute.amazonaws.com:8000/info

View info about the underlying model

curl ec2-54-183-245-15.us-west-1.compute.amazonaws.com:8000/predicts

Make a prediction using data

curl ec2-54-183-245-15.us-west-1.compute.amazonaws.com:8000/predicts -L -X POST -d '{"sepal_length": [6.9], "sepal_width": [3.2], "petal_length": [5.7], "petal_width": [2.3]}' -H 'Content-type: application/json'

Prediction Response: "{'Species': 'Iris-virginicia'}"

If deployed locally, you can try it out with the following (note: 0.0.0.0 is significant, localhost sometimes binds to 127.0.0.1 and may not work if the model is running within the datmo container):

curl 0.0.0.0:8000/predicts -L -X POST -d '{"sepal_length": [6.9], "sepal_width": [3.2], "petal_length": [5.7], "petal_width": [2.3]}' -H 'Content-type: application/json' 

Built With

• Datmo - Model versioning and env builder
• Gunicorn - Python WSGI HTTP server framework
• Falcon - Highly performant baremetal Python API Framework that compiles in C