Titanic Lightgbm Prediction Webservice

Aim

• To build a web service that accepts input, runs a prediction on it, and returns the predicted result with probability using the lightgbm model on Titanic data and implementing a synchronous API and an asynchronous API approach.

Building the webservice

• The standard library’s virtual environment tool venv, was used to develop the service in a virtual environment and install packages.
• A list of all installed packages and their versions can be found in the requirements.txt file.
• Titanic data was obtained from Kaggle. Subsequent steps for loading, preprocessing, analysing and transforming data are carried out in the model.py file
• LightGBM - was installed to implement the prediction model in the model.py file. In addition, the model was exported to be used in the sync and async API.
• FastAPI - was used to set up sync and async endpoints.
• Logging - was used to track the web service performance.
• Celery - an open-source asynchronous job queue based on distributed message passing, supports scheduling and focuses on operations in real time, was used to implement the async api using Redis as a message broker to run in the async.py file
• The sync api endpoint implementation can be found in the sync.py file

Requirements

Docker

Techstack

• LightGBM
• FastAPI
• uvicorn
• Celery
• OpenAPI
• SwaggerUI
• Redis
• Docker

Running the web service

• Uvicorn - was used to serve the web service and containerised in Dockerfile.sync and Dockerfile.async, respectively. To load the server for sync and async endpoints, run the following commands in the console:

docker run -p 8000:8000 lgb-fastapi-sync-app

docker run -p 8000:8000 lgb-fastapi-async-app

• If everything works as intended, you can see the server running at http://127.0.0.1:8000.

Using the web service

• After starting your server, you can run the following scripts for sync and async endpoints and provide a JSON input to see if the web service works as intended:

sync_testing.py

async_testing.py

Developer

Jehoram Mwila, Data and MLOps Engineer