PredictionAPI

This is a tutorial on serving predictions from a machine learning model through a REST API.

A version of this tutorial was presented in the PyParis conference on 13/06/2017.

A complete written version is currently in the making.

Steps

training and persisting the model on disk (cf. modeling.ipynb),
reproducing the pre-processing pipeline in the API application (cf. application.py),
encapsulating the application in a Docker container (cf. Dockerfile),
deploying the application to a cloud server.

Technical Requirements

Python 3.5+ and virtualenv,
Docker,
The required Python librairies used can be installed from the included requirements.txt file:

virtualenv -p python3 pyenv
source pyenv/bin/activate
pip install -r requirements.txt

Running the application locally

Directly

cd PredictionAPI
export FLASK_APP=application.py
python3 -m flask run

On Docker

cd PredictionAPI
docker build -t prediction-api .
docker run -d --rm -p 5000:5000 prediction-api

Testing the application

Once it is running, the API can be queried using HTTP POST requests. This can be done from the CLI using curl or through a GUI REST client like Insomnia or Postman.

URL: http://0.0.0.0:5000/api/v1.0/aballone

Here is a sample query:

{    
  "inputs": [
    {
      "sex":"M",
      "length": 0.815000,
      "diameter": 0.055000,
      "height": 1.130000,
      "whole_weight": 2.825500,
      "shucked_weight": 1.488000,
      "viscera_weight": 0.760000,
      "shell_weight": 0.001500
    },
    {
      "sex":"F",
      "length": 0.815000,
      "diameter": 1.055000,
      "height": 1.130000,
      "whole_weight": 2.825500,
      "shucked_weight": 1.488000,
      "viscera_weight": 1.760000,
      "shell_weight": 0.001500
    }
  ]
}

The response should look like this:

{
  "outputs": [
    {
      "label": 1,
      "prob": 0.109
    },
    {
      "label": 1,
      "prob": 0.183
    }
  ]
}

License

NawfalTachfine/PredictionAPI

ErrorLooks like something went wrong!

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases 1

Packages 0

Uh oh!

Languages

Error
Looks like something went wrong!

Packages