Freshwater Hybrid Modelling

Variational Graph Auto-Encoder to embed Latex equations

This repository contains the code to preprocess, train and test Latex equations with a VGAE.

Environment Setup

1. Clone the project
2. Create a dataset and trained_models within the root folder
3. Setup a virtual environment in the root folder called venv (more on that here):
   • Download python version 3.10.10
   • Create a new environment with python -m venv venv.
   • Activate the virtual environment with the command: source venv/bin/activate.
   • Make sure the Python version is 3.10.10 with python -V.
4. Install the librairies with pip install -r requirements.txt.

Code Architecture

The code is split into 4 different folders:

1. Node contains the API to transform Latex equations into MathML
2. Preprocessing takes care of processing the dataset of equations and build a big XML file, a vocabulary and the Graph Dataset
3. Models involves the files to train, do hyperparameter search, and also the VGAE model
4. Utils contains the code to plot, save, and extract experiment data

All these files can be called through main.py in the root folder. The training parameters are held within the config.py file.

Authors

• Nicolas SAMELSON - https://github.com/nsamelson