An english to french Machine Translation Web Tool - powered by an RNN (LSTM based Seq2Seq Model)

Install

• when opening VSCode, install the suggested extensions (Python, Black Formatter and Pylance)
• create your python environment python3 -m venv .venv. Make sure to have a python version that tensorflow supports (ex. 3.11.4)
• use Gitbash terminal
• activate your environment with source activate path\to\activation\file
• run the server with python app.py

The server should answer on http://localhost:5000

You can deactivate the environment with deactivate.

Adding librairies

If you need to use new librairies, you can do it with pip pip install [library name] or pip3 install [library name] or 'conda install'

The Project

• This is an academic project in the context of a "introduction to software development" course at Ecole des Ponts et Chaussées.
• The dataset exists in the folder Dataset and was taken from Kaggle open datasets.
• The provided python notebook contains data preprocessing and vectorization, model definition and training and inference. It's the main part of the project.
• The web interface is developed using python (flask toolkit) for backend and html/js/css for frontend.

Requirements :

• Actually, not all the packages in the file requirements.txt are required for this project. This file will be soon updated.

N.B:

This is my 1st NLP project. So it means something to me.
It was trained on a single T4 GPU for less than 1h, on a vocabulary size of around 10K and a maximum sequence length of 20.
Hence, you shouldn't expect it to perform very well eventhough it showed descent results 😁.
I will add a transformer-based new version and scale compute and data budget.