README.md

Table of Contents

• Getting Started
• Prerequisites
• Installing
• Running the App
  • Local

Getting Started

• This project is about generating text from image. To get started, follow the instructions below.
• This is the project structure:

|-jenkins                      - Directory for Jenkins configuration files
|-k8s                          - Directory for Helm chart to deploy the application
|-models                       - Directory for Deep Learning models
|--vocabs.json                 - File for vocabulary NLP model
|-notebooks                    - Directory for crawling and training data
|--data                        - Directory for data (crawling)
|--data_crawler.ipynb
|--requirements.txt            - File contains packages for training and crawling
|--training.ipynb
|-src                          - Directory for main application
|-Dockerfile                   - File to create and deploy main application
|-Makefile                     - File contains some commands (make + <cmd>)
|-requirements.txt             - File contains packages for main application

Prerequisites

• Git
• Anaconda (optional)
• Python: 3.9
• Kubernates
• Minikube (for weak machine) - using this in this project or Minikf
• Docker
• Makefile (optional)

Installing

Using anaconda:

conda create -n <env_name> python=3.9
conda activate <env_name>
pip install -r requirements.txt

Running the App

Local

Running in Local

uvicorn src.main:app --reload

Or

make start-app-dev

Running in Docker

docker build --tag "jigglediggle1/image-captioning" .
docker run -d --name="image-captioning" -p 30000:30000 "jigglediggle1/image-captioning"

Or

make start-app-dev

Running with K8S

Start the Docker (Engine or Desktop)

Notes: The docker's memory default only has 3GB memory, may not enough for this project, should increase it (in Desktop) to at least 6G

Start the Minikube

Notes: The minikube's memory default only has 2GB memory, may not enough for this project, should increase it by the following command in MB (and it should be lower than Docker's memory)

minikube config set memory 4000

And delete the minikube (if you already initialized one before).
WARNING: This will delete all of your pods

minikube delete

Then start the minikube

minikube start

Minikube has dashboard to monitor the pods and more. To open it, run:

minikube dashboard

Run the K8S Helm Chart

helm upgrade --install nginx-ingress ./k8s/nginx-ingress/. --namespace=image-captioning
helm upgrade --install ic-app ./k8s/ic-app/. --namespace=image-captioning

Then, we will need to run minikube tunnel

minikube tunnel

And finally, open new terminal, run port-forward to use the application

kubectl --namespace image-captioning port-forward svc/ic-app 5000:30000

Or

export NODE_PORT=$(kubectl get --namespace {{ .Release.Namespace }} -o jsonpath="{.spec.ports[0].nodePort}" services {{ include "ic-app.fullname" . }})
kubectl --namespace {{ .Release.Namespace }} port-forward $POD_NAME 5000:30000

And open this url to visit the application: localhost:5000/docs