Skip to content

Repository for the project of Machine Learning Operations course at DTU. The aim of the project is to put in practice the principles of MLOps and get familiarized with the best tools for it.

License

Notifications You must be signed in to change notification settings

dibuja/mlops-project

Repository files navigation

MLOps Project: Classification of Yelp Business Reviews Based on Sentiment Analysis

The Project

This repository contains a project conducted within the scope of the DTU's course "Machine Learning Operations" which can be found here.

The aim of the project is to learn the principles of MLOps and put in practice the best techniques and methods for it.

The structure of the project follows the cookie cutter structure.

The topic of the project consists of classifying reviews of businesses and try to predict the number of stars a review has based on the sentiment analysis of the comments left by its author. For that purpose, the BERT language model taken from the Transformers repository is used.

The following features have been implemented:

  • Using the cookie cutter structure
  • Pre-commits with code formatting and checking
  • Separation of the algorithms in make_dataset, model file and training scripts
  • Requirements.txt file
  • Dedicated Conda environment for the project
  • Compliance with pep8 coding conventions
  • Usage of Google Cloud Platform for data storage, model deployment
  • Experiment testing using Hydra
  • pre-commit hooks
  • Data version control with DVC
  • Monitoring using Wandb

Authors

About

Repository for the project of Machine Learning Operations course at DTU. The aim of the project is to put in practice the principles of MLOps and get familiarized with the best tools for it.

Topics

Resources

License

Stars

Watchers

Forks

Contributors 3

  •  
  •  
  •