arnauddelaunay / apiwrapper

Wrap your Machine Learning model into a Flask API.

API WRAPPER

ApiWrapper wrap your ML model into an API. Default routes are predict and predict_proba.

Install

git clone https://github.com/arnauddelaunay/apiwrapper.git
cd apiwrapper
sudo pip install -r requirements.txt

Usage

Once your model is loaded and fit, you can execute the api and pass the model as an argument

#!python
import apiwrapper
API = apiwrapper.Api(model=model)
API.run(port=5000, debug=True)

Endpoints

• / [GET] : info about the model
• /predict [POST] : return the prediction of the model for the given input. FORMAT : {"data" : np.ndarray}
• /predict_proba [POST] : return the prediction probabilities of the model for the given input. FORMAT : {"data" : np.ndarray}

Complete exemple

Run python main.py

And test the API :

curl -X POST -H "Content-Type: application/json" -d '{"data" : [
	[0.1, 0, 0.6, 1.2],
	[ 1.2,  1.3,  3.1, 1.8]
	]
}' "http://localhost:5000/predict"

gives the following results : {"results": [0,2]}.

DOCKERIZE THE API

Use Docker to serve your API in background on the local network.

Build

Go into the root of this repo

$ sudo docker build -t mymodel .

Run

$ sudo docker run -d --name mycontainer -p 5000:5000 mymodel

Test

Check your docker IP (here 172.17.0.1):

$ ifconfig
docker0   Link encap:Ethernet  HWaddr 02:42:a8:32:08:90  
          inet adr:172.17.0.1  Bcast:0.0.0.0  Masque:255.255.0.0
          ... ... ...
          ... ... ...

You can now test the API :

$ curl -X POST -H "Content-Type: application/json" -d '{"data" : [
	[0.1, 0, 0.6, 1.2],
	[ 1.2,  1.3,  3.1, 1.8]
	]
}' "http://172.17.0.1:5000/predict"

gives the following results : {"results": [0,2]}.