https://github.com/singhdivyank/sweet-sense-ai

Integrate binary classifier in a website to predict diabetes. Tech: XG Boost, AWS ECS, JavaScript
https://github.com/singhdivyank/sweet-sense-ai

aws-ecs-fargate docker exploratory-data-analysis fastapi javascript supervised-machine-learning xgboost-classifier

Last synced: 12 days ago
JSON representation

Integrate binary classifier in a website to predict diabetes. Tech: XG Boost, AWS ECS, JavaScript

• Host: GitHub
• URL: https://github.com/singhdivyank/sweet-sense-ai
• Owner: singhdivyank
• Created: 2024-12-23T20:30:01.000Z (17 days ago)
• Default Branch: main
• Last Pushed: 2024-12-24T05:29:16.000Z (16 days ago)
• Last Synced: 2024-12-24T06:22:45.467Z (16 days ago)
• Topics: aws-ecs-fargate, docker, exploratory-data-analysis, fastapi, javascript, supervised-machine-learning, xgboost-classifier
• Language: Jupyter Notebook
• Homepage:
• Size: 896 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: readme.md

Awesome Lists containing this project

README

        
# Objective

Turn a binary classifier (**XG Boost**) to predict diabetes after Exploratory Data Analysis into a Machine Learning Solution.

# Architecting the solution

The solution is intended to use a Machine Learning model to power a web application and is architected in two stages - i) model pipeline, ii) integrating into a website

## Model pipeline

A three task strategy

Table1: model deployment pipeline


| S. No. | Task | Purpose | Tech |

| ------ | ---- | ------- | ---- |

| 1 | API | Model processes a request and sends a response | FastAPI |

| 2 | Containerization | wrapper that captures all dependencies | Docker |

| 3 | Deployment | run the container over a server or cloud provider | AWS |

Steps associated with the model deployment pipeline:

1. Create RestAPI to recieve parameters and return the predictions

2. Containerize the API into a [Docker image](https://fastapi.tiangolo.com/deployment/docker/) and push to Docker hub

3. Deploy the container on AWS ECS

### Executing FastAPI

API details

1. method: POST

2. endpoint: `/inputs`

3. body: glucose (float), insulin (float), bmi (float), age (int)

4. response: dict {"status": "diabetic/undiabetic"}

Follow the steps below to create and execute the RestAPI on your local server and the [Postman Desktop Client](https://learning.postman.com/docs/getting-started/first-steps/get-postman/). 

1. `cd app`

2. `pip install -r requirements.txt`

3. `python3 -m uvicorn main:app --server 127.0.0.1 --port 8001 --reload`



![](img/apiExecution.png)



Figure1: API execution results


4. exit Uvicorn server once everything is working and build Docker image

    

    a. `cd ..`

    

    b. `docker build -t  .`

5. run the docker image, `docker run -d --name  -p 80:80 `

6. create a repository on [docker hub](https://hub.docker.com) and push the image

    a. `docker tag  `

    b. `docker push `

7. query the endpoint on Postman



![](img/apiExecution_Docker.png)



Figure2: executing Docker image endpoint on Postman







![](img/DockerExec.png)



Figure3: Docker image execution status


## Website Integration

### On a side note: an alternate approach

Creating a website in JavaScript and integrating the ML classifier can be avoided by using **no-code** alternatives such as [Gradio](https://www.gradio.app). As a no-code platform it creates an API to access the model from an iteractive UI. This allows one to access the model directly as a pickle file which helps avoid both containerization and deployment.

This implementation can be found in `gradio.py`. Steps to execute on Gradio

1. `pip install gradio==3.36.1`

2. `python3 gradio.py`

**Note**: FastAPI is a Gradio dependency. An error may arise due to version conflict between both of them. Just search for the correct versions and install them.