A sample ML app that trains a model and we deploy it to k8s

The project contains a ml app Machine Learning application, and the docker configuration to run it locally or deploy it on k8s platform.

• The root directory of the project contains the following files
  • ml-app-k8s folder that contains all the spring boot code base
    • charts folder contains helm chart files.
      • templates folder contains the following:
        • deployment.yaml contains K8s deployment specifics and env level variables.
        • hpa.yaml is used to specify auto-pod scaling in the cluster
        • NetworkPolicy.yaml file manages external access to the services in a cluster, typically HTTP.
        • pdb.yaml file is used to manage Pod Distribution Budget that the cluster will honor
        • service.yaml specifies service management like the loadbalancer with AWS
        • cronjob.yaml CronJob manifest prints the current time and a simple hello message every 5 minute
      • chart.yaml is used to specify app-pod level information that is used throughout the K8s config
      • values.yaml is used to handle app information and specifics. we can use values/dev.yaml to override these values and use them for each environment.
  • core folder is used to importing the packages we need, as well as the classes, which will be used to train our models
  • data folder is used to contain the datasets based out of Auto MPG dataset which is used for deploying a machine learning model
  • models folder contains the model which is dumped and exported to a file, that we can load into Python at another point in time
  • main.py contains the code to expose the model as a service
  • train.py contains the code to train our model by specifying which feature we want to predict
  • Dockerfile is the dockerfile used to build the image and/or run the app

To run this project locally, please cd to the location this template is present and run the following:

docker docker build --tag ml-app .
docker run -p 8080:8080 -itd ml-app:latest

deploy the application to your kubernetes cluster using HELM.

kubectl config set-context `clusterName` --namespace `namespace`

cd chart/ml-app

helm upgrade ml-app .

now lets fetch the external or internal endpoint of the service that is deployed

kubectl get svc

We have the external-ip from earlier, which we are going to reuse here. The following JSON request can now be sent as with a HTTP POST method, and we will receive the expected response.

{
    "cylinders": 8,
    "displacement": 307.0,
    "horsepower": 130.0,
    "weight": 3504,
    "acceleration": 12.0,
    "model_year": 70,
    "origin": 1,
    "car_name": "chevrolet chevelle malibu"
}

In just 0.29 seconds, we received the prediction of this car, and our machine learning model predicted 16.383.