Project Lifecycle and setup

This repository hosts a machine learning model deployed as a RESTful API using FastAPI. It leverages a RandomForestClassifier model trained on NY building data, offering an endpoint for making predictions.

Project tree

.fastapi-model-deploy

• train_app
  • retrain.py - training/re-training
  • README.md - Pre-processing findings
• infra_app
  • app.py - Flask API file
  • incomemodel.py - pydantic for input validation
  • Dockerfile - serving container
  • download_latest_model.py - Download latest model from mlflow
  • requirements.txt
  • README.md - Production style pipeline
• data
  • train - training data
  • test - testing data
• notebooks -- Data analysis/preprocessing
• models -- save trained model to this dir
• README.md
  • Instructions for setting up and running the repo locally

Prerequisites

• python 3.9+
• pip
• docker

Instruction

1. Clone the repo
2. Create a virtual env
3. Install the dependencies for model training in train_app
   • pip install -r requirements.txt
4. run 'retrain.py' --> 'python retrain.py'
   • mlflow will be up and running locally and the new model that we currently trained will be register
   • navigate to the Mlflow url to check out
   • You will see the model saved in models directory
5. Now move to infra_app directory
   • run 'download_latest_model.py'
   • this will connect to mlflow get the latest model and saves in the pwd
6. Run Dockerfile to build the image
   • 'docker build -t flask-api .'
   • 'docker run -p 8000:8080 flask-api'
7. Test the api
   • connect to api docs and test --> http://localhost:8000/docs