Housing Prices ML Regression API

Machine Learning Engineer Project
Author: Mateus Cichelero
Date: February 2024

This project provides a REST API that uses a machine learning regression model to predict housing prices for the Sound Realty Company. The API allows users to input housing characteristics and receive a price prediction, supporting Sound Realty in their valuation and decision-making processes.

Technologies

Python: The primary programming language.
FastAPI: Framework for rapidly building web APIs.
Scikit-Learn: Tools for machine learning, including the regression model.
Pandas: Library for data manipulation and analysis.
Poetry: For managing Python dependencies.
Docker: Containerization for easy development and deployment.

Features

Endpoint /predict: Accepts comprehensive housing data and returns a price prediction.
Endpoint /predict_minimal: Provides predictions using a core set of features.
Data Validation: Ensures input data aligns with model expectations.
Swagger Documentation: Automatic API documentation for ease of use.

Challenge Requirements

Deploy the model as an endpoint on a RESTful service which receives JSON POST data. (see /predict endpoint at app/main)
Create an additional API endpoint where only the required features have to be provided in order to get a prediction. (see /predict_minimal endpoint at app/main)
Create a test script which submits examples to the endpoint to demonstrate its behavior. (try cd experiments and then python experiments/api_test_script.py in your poetry env with the api application already running on docker)
Evaluate the performance of the basic model. (see regression metrics comment at #11 - use of cml from iterative (https://cml.dev/). Implemented at experiments/create_model.py in experiments-basic-model branch)
Improve the model by applying some basic machine-learning principles. Check the implementation of catboost regressor with grid search and using categorical features in the experiments-gbm-model branch. See regression metrics comment at #12 (a good improvement compared to the basic model)

Basic Usage:

Create environment with Poetry

With poetry alread installed and at the root of this repository:

> poetry install
> poetry shell

after having the env ready, you can run the API locally with:

uvicorn app.main:app --reload

Using the API with Docker

With docker engine already installed:

> docker build -t phdata-api .
> docker run -d --name mycontainer -p 80:80 phdata-api

After that you should be able to access the API docs from http://0.0.0.0/docs (and to apply the test script from experiments/api_test_script.py)