MLOps Credit Default ✈️

This is a personal MLOps project based on a Kaggle dataset for credit default predictions.

It was developed as part of the this End-to-end MLOps with Databricks course and you can walk through it together with this Medium publication.

Feel free to ⭐ and clone this repo 😉

Tech Stack

Project Structure

The project has been structured with the following folders and files:

• .github/workflows: CI/CD configuration files
  • cd.yml
  • ci.yml
• data: raw data
  • data.csv
• notebooks: notebooks for various stages of the project
  • create_source_data: notebook for generating synthetic data
    • create_source_data_notebook.py
  • feature_engineering: feature engineering and MLflow experiments
    • basic_mlflow_experiment_notebook.py
    • combined_mlflow_experiment_notebook.py
    • custom_mlflow_experiment_notebook.py
    • prepare_data_notebook.py
  • model_feature_serving: notebooks for serving models and features
    • AB_test_model_serving_notebbok.py
    • feature_serving_notebook.py
    • model_serving_feat_lookup_notebook.py
    • model_serving_notebook.py
  • monitoring: monitoring and alerts setup
    • create_alert.py
    • create_inference_data.py
    • lakehouse_monitoring.py
    • send_request_to_endpoint.py
• src: source code for the project
  • credit_default
    • data_cleaning.py
    • data_cleaning_spark.py
    • data_preprocessing.py
    • data_preprocessing_spark.py
    • utils.py
• tests: unit tests for the project
  • test_data_cleaning.py
  • test_data_preprocessor.py
• workflows: workflows for Databricks asset bundle
  • deploy_model.py
  • evaluate_model.py
  • preprocess.py
  • refresh_monitor.py
  • train_model.py
• .pre-commit-config.yaml: configuration for pre-commit hooks
• Makefile: helper commands for installing requirements, formatting, testing, linting, and cleaning
• project_config.yml: configuration settings for the project
• databricks.yml: Databricks asset bundle configuration
• bundle_monitoring.yml: monitoring settings for Databricks asset bundle

Project Set Up

The Python version used for this project is Python 3.11.

1. Clone the repo:

   git clone https://github.com/benitomartin/mlops-databricks-credit-default.git

2. Create the virtual environment using uv with Python version 3.11 and install the requirements:

    uv venv -p 3.11.0 .venv
    source .venv/bin/activate
    uv pip install -r pyproject.toml --all-extras
    uv lock

3. Build the wheel package:

   # Build
   uv build

4. Install the Databricks extension for VS Code and Databricks CLI:

   curl -fsSL https://raw.githubusercontent.com/databricks/setup-cli/main/install.sh | sh

5. Authenticate on Databricks:

   # Authentication
   databricks auth login --configure-cluster --host <workspace-url>
   
   # Profiles
   databricks auth profiles
   cat ~/.databrickscfg

After entering your information, the CLI will prompt you to save it under a Databricks configuration profile ~/.databrickscfg

Catalog Set Up

Once the project is set up, you need to create the volumes to store the data and the wheel package that will you have to install in the cluster:

• catalog name: credit

• schema_name: default

• volume name: data and packages

  # Create volumes
  databricks volumes create credit default data MANAGED
  databricks volumes create credit default packages MANAGED
  
  # Push volumes
  databricks fs cp data/data.csv dbfs:/Volumes/credit/default/data/data.csv
  databricks fs cp dist/credit_default_databricks-0.0.1-py3-none-any.whl dbfs:/Volumes/credit/default/packages
  
  # Show volumes
  databricks fs ls dbfs:/Volumes/credit/default/data
  databricks fs ls dbfs:/Volumes/credit/default/packages

Token Creation

Some project files require a Databricks authentication token. This token allows secure access to Databricks resources and APIs:

1. Create a token in the Databricks UI:

   • Navigate to Settings --> User --> Developer --> Access tokens

   • Generate a new personal access token

2. Create a secret scope for securely storing the token:

   # Create Scope
   databricks secrets create-scope secret-scope
   
   # Add secret after running command
   databricks secrets put-secret secret-scope databricks-token
   
   # List secrets
   databricks secrets list-secrets secret-scope

Note: For GitHub Actions (in cd.yml), the token must also be added as a GitHub Secret in your repository settings.

Now you can follow the code along the Medium publication or use it as supporting material if you enroll in the course. The blog does not contain an explanation of all files. Just the main ones used for the final deployment, but you can test out other files as well 🙂.