Introduction#

Author : Cyril Achard - 310048

Jupyter Book structure#

This book contains the Homework 2 of the Deep Learning in Biomedicine course.

  • The first section introduces the dataset and the preprocessing used.

  • The second section contains reports of hyperparameter tuning for the different models.

Warning

The reports are embedded html reports from Weights and Biases. If you encounter any problem, please use the provided links to access the reports at the bottom of each section instead.

  • The third and last section contains the notebooks with the best run of each model executed.

Table of contents#

Dataset & preprocessing

Hyperparameter tuning

Best models notebooks

Project structure#

The project directory is structured as follows:

  • WEBSITE : The built Jupyter Book, containing the notebooks and analysis

  • report : Contains the PDF report

  • book_src : Contains the Jupyter Book source data and all notebooks/code

    • code : Source code of the project as .py files

      • model.py : Contains the models code (layers, heads, aggregators, etc.)

      • training.py : Contains the training code (training loop, evaluation, etc.)

      • utils.py : Contains data-loading/pre-processing and plotting utilities

    • rendered_notebooks : Rendered notebooks of best runs
      (See Graph Convolutional Network)

    • wandb_comparisons : HTML reports of the hyperparameter tuning
      (See Hyperparameter tuning for GCN)

    • images : Images used in the book and report

Tools used#

  • Hyperparameter tuning and interactive plots with Weights and Biases

  • Models are pure PyTorch (geometric was not used, even for data loading)

  • Data loading with HuggingFace datasets

  • Graphs visualisation with NetworkX

  • Documentation and structured notebooks with Jupyter Book

  • Report with Overleaf

  • Code formatting with pre-commit and ruff (w/ black and isort)