ML Example App

An Example of how to deploy ML model and scale with kubernetes

Requirements: Docker & Kubernetes

Ensure that your kubectl is connected to a kubernetes cluster This can be done by modifying your kubectl config with a system path pointing to KUBECONFIG

Private Docker Registry

You must have a docker registry for the docker images to be sent to. indicate the host to your docker registry in the file docker-registry-address

Local kubernetes setup

If you dont have a kubernetes cluster, try microk8s And then install kubectl separately:

Configure your kubectl with

cat microk8s.config >> ~/.kube/config Then add export KUBECONFIG=~/.kube/config into your bashrc

Installing Docker Registry

On microk8s, enable docker registry with microk8s.enable registry

Allowing docker to push to private docker registry

Add the address of the docker private registry into your insecure-registries E.g http://localhost:32000 For Mac/Windows:

open the settings, goto the daemon tab and then pop in your registry’s URL in the “Insecure registries” Restart docker

For Ubuntu:

vim /etc/docker/daemon.json

{
    "insecure-registries" : ["localhost:32000"]
}

Restart your docker service with systemctl restart docker

Installation instructions

Just run the included install.sh file

sh install.sh

Removal

Just run the included uninstall.sh file

sh uninstall.sh

Features

Frontend --> UI built with ReactJS to request predictions

Backend --> Backend built with Python Flask for

     sanitize inputs
     Track & record predictions
     REST API for model serving
     Requests Model Serving API
     REST API for past predictions

Mushroom Model --> ML Pipeline built with MLFlow

   Website for viewing model training / track models
   WorkFlow for Training and deploying model
   Model trained as a container
   Model deployed as an isolated container

Nasdin/Example-ML-CICD-fullstack