A classification model on publicly available Census Bureau data, deployed using FastAPI & Render, with CI/CD incorporated using GitHub Actions.
The main files of this repository follow the following structure.
.
├── .dvc/ # DVC Configs
├── .github/workflows/ # CI/CD (with Github Actions) Codes
├── data/ # Metadata for Input Data (DVC Enabled)
├── model/ # Metadata for Model-Related Files (DVC Enabled)
├── notebooks/ # Jupyter Notebooks
│ ├── EDA.ipynb
│ ├── Model Building.ipynb
│ ├── Model Bias & Fairness.ipynb
├── screenshots/ # Screenshots for Documentation
├── src/ # ML Pipeline Codes
│ ├── __init__.py
│ ├── data_cleaning.py
│ ├── model_runner.py
│ ├── utils.py
├── tests/
│ ├── __init__.py # Unit Tests for FastAPI App & ML Pipeline
│ ├── test_app.py
│ ├── test_ml_pipeline.py
├── main.py # API Codes
├── render_api_request.py # live API POST method
├── sanitycheck.py # API Checking
├── setup.py
├── requirements.txt
├── slice_output.txt
├── xgboost_model_card.md
└── README.md
To install the requirements, run the following command:
pip install -r requirements.txt
To run the app locally, run the following command:
uvicorn main:app --reload
To run the unit tests, run the following command:
pytest tests/ -vv
To run individual scripts inside src/, run the following command:
python src/{SCRIPT_NAME}.py
CI (Continuous Integration) and CD (Continuous Deployment) is integrated to this repository through the use of GitHub Actions.
After each commit to the main branch, GitHub Actions will initiate the build job, check for syntax errors using Flake8, and run a series of tests (refer to scripts in tests/) using Pytest. These processes will serve as the Continuous Integration (CI) step before deployment.
Once the build process is completed (and all tests have been passed), the deployment job will start. This will ensure Continuous Deployment (CD) to the Render cloud app.
For more information on GitHub Actions, refer to the GitHub Actions documentation.
An XGBoost model was trained on the Census Bureau Dataset, obtained from the UCL ML Repository, to predict whether a person's annual income is over or under $50K. More details on the model (as well as the EDA, training, and model fairness assessment process) can be found in the model card.
The data and model-related (label encoder, scaler, and model pkl) files are version-controlled using DVC. The metadata for these files can be found inside data/ and model/.
For more information on DVC, refer to the DVC documentation.
The API is built using FastAPI, using the codes found in main.py. A snippet of the automatically-generated documentation by FastAPI can be found below.
For API deployment, Render is used. A snippet of the live page receiving the contents of the GET method can be found below.
The POST method can be used to obtain a prediction from the trained ML model. To do this, input the desired feature values inside the render_api_request.py file. Then, run the following command:
python render_api_request.py
An example of this process, and the response generated from the live API, can be found below.
For more information on FastAPI and Render, refer to the FastAPI and Render documentations.