Kirtan001/R_Bank_Marketing_UCI_Classification

End-to-End MLOps: UCI Bank Marketing Analysis

This project is a complete MLOps implementation for predicting term deposit subscriptions using the UCI Bank Marketing dataset. It includes data processing, model training (XGBoost), experiment tracking (MLflow), and deployment (Shiny App & Plumber API).

📂 Project Structure

├── analysis
│   └── (EDA and research files)
├── data
│   └── (Dataset files downloaded automatically)
├── src
│   ├── app.R          # Interactive Shiny App for Hugging Face
│   ├── plumber.R      # REST API for model serving
│   └── model.rds      # Trained model artifact (generated by report)
├── tests
│   └── (Unit tests for model validation)
├── .github/workflows
│   └── main.yaml      # CI/CD: Builds & Pushes Docker Image to GHCR
├── Dockerfile         # Configuration for containerization
├── final_report.Rmd   # Main report: Data -> Training -> Evaluation
└── README.md          # Project documentation

🚀 How to Run

1. Generate the Model

Before running the app or API, you must train the model. Open final_report.Rmd in RStudio and click Knit.

• —This will run the entire pipeline.
• —It protects against data leakage.
• —It saves the best model to src/model.rds.

2. Run Locally (Shiny App)

Open src/app.R in RStudio and click Run App.

3. Run Locally (Plumber API)

Run the following R code:

library(plumber)
pr <- plumb("src/plumber.R")
pr$run(port = 8000)

☁️ Deployment (Hugging Face Spaces)

This project is configured for Docker deployment on Hugging Face.

Option A: Direct GitHub Connection (Recommended)

1. 1.Push to GitHub: Commit all files (including src/model.rds) and push to your repository.
2. 2.Create Space: Go to Hugging Face Spaces -> "Create new Space".
3. 3.Select SDK: Choose Docker.
4. 4.Connect Repo: Select your GitHub repository.
5. 5.Build: Hugging Face will automatically build the Docker image and launch the app.

Option B: Use Pre-Built Image from GitHub Packages (GHCR)

The CI/CD pipeline automatically builds and pushes the Docker image to GitHub Container Registry.

1. 1.Go to your GitHub Repository -> Packages.
2. 2.Copy the full image URL (e.g., ghcr.io/username/repo:main).
3. 3.In Hugging Face Space settings, you can configure it to pull this image (requires Access Token if private).

Option C: Automated CI/CD (GitHub Actions) [Recommended]

We have configured main.yaml to automatically push code improvements to Hugging Face.

1. 1.Get Token: Go to Hugging Face -> Settings -> Access Tokens -> Create "Write" Token.
2. 2.Add Secret: Go to GitHub Repo -> Settings -> Secrets and variables -> Actions -> New Repository Secret.
3. 3.Name: HF_TOKEN
4. 4.Value: (Paste your token)
5. 5.Deploy: Any push to main will now auto-update your Space!

🤖 CI/CD Pipeline

The main.yaml workflow performs the following:

1. 1.Lints Code: Checks R syntax for errors.
2. 2.Builds Docker Image: Creates the production container.
3. 3.Pushes to GHCR: Stores the image in GitHub Packages for easy deployment.

🛠️ Technologies

• —Modeling: tidymodels, xgboost, ranger, themis (SMOTE).
• —Deployment: shiny, plumber, docker.
• —Validation: testthat, yardstick.