An example of deploying a sklearn model using Flask using a Docker container.

This tutorial requires basic Docker knowledge.

For this example we are training a toy model using Iris training dataset. To train a new model, run this:

python train.py

This outputs a pickle model in a file named model.pkl.

Construct an image (docker build) called chrisalbon/sklearn-flask-docker (--tag chrisalbon/sklearn-flask-docker) from the Dockerfile (.).

The construction of this image is defined by Dockerfile.

docker build --tag chrisalbon/sklearn-flask-docker .

Create and start (docker run) a detached (-d) Docker container called sklearn-flask-docker (--name sklearn-flask-docker) from the image chrisalbon/sklearn-flask-docker:latest where port of the host machine is connected to port 3333 of the Docker container (-p 3000:3333).

docker run -p 3000:3333 -d --name sklearn-flask-docker chrisalbon/sklearn-flask-docker:latest

Since our model is trained on the Iris toy dataset, we can test the API by queries it for the predicted class for this example observation:

• sepal length = 4.5
• sepal width = 2.3
• petal length = 1.3
• petal width = 0.3

Paste this URL into your browser bar:

http:https://0.0.0.0:3000/api/v1.0/predict?sl=4.5&sw=2.3&pl=1.3&pw=0.3

In your browser you should see something like this:

{"features":[4.5,2.3,1.3,0.3],"predicted_class":0}

"predicted_class":0 means that the predicted class is "Iris setosa"

Paste this URL into your terminal:

curl -i "0.0.0.0:3000/api/v1.0/predict?sl=4.5&sw=2.3&pl=1.3&pw=0.3"

You should see something like this:

HTTP/1.0 200 OK
Content-Type: application/json
Content-Length: 51
Server: Werkzeug/1.0.1 Python/3.6.12
Date: Tue, 25 Aug 2020 20:29:41 GMT

{"features":[4.5,2.3,1.3,0.3],"predicted_class":0}

You will need to use sudo for these commands, however best practice is to add your user to the docker group when in production.

docker start sklearn-flask-docker

docker stop sklearn-flask-docker

docker rm sklearn-flask-docker