namnd00 / ml-pipeline-deploy-on-render

Guide in this project: ML pipeline to expose API on Render

Introduction:

This project to create a pipeline to train a model and publish it with a public API on Render.

Development:

• Install nescessary packages: pip install -r requirements.txt
• Build Machine Learning pipeline and you can run full end to end pipeline on script:python ml_pipeline.py in the following help:

usage: ml_pipeline.py [-h] [--steps {basic_cleaning,train_model,evaluate_model,all}]

ML training pipeline

optional arguments:
  -h, --help            show this help message and exit
  --steps {basic_cleaning,train_model,evaluate_model,all}
                        Pipeline steps

• The prepared data saved to path and trained model, other artifacts saved to path

Testing

• Running test case on script: pip install -e . && pytest . to install starter module and execute pytest script.

API

• Running script: uvicorn main:app --host 0.0.0.0 --port 10000 and access to link to using swagger UI and try to use API.

CI/CD

Note: I've used google drive to save data, model instead AWS S3.

Pipeline:

• Pulls data from DVC google drive, sync model to deploy directory and execute Flake8 + pytest doing every test.
• Automatically deploy to production on Render cloud platform instead Heroku
• This deployed API link: https://mlops-app.onrender.com/

Check Render deployed API

• Running script: python check_render_api.py
• The stdout should be:

Response code: 200
Response body: {'prediction': '[0]', 'class_name': '<=50K', 'success': True}

Screenshots

• You can access to screenshots to see them.