Environment setup

The project uses uv as python package management, follow the official guide to install it.

# Then, download all the required packages
uv sync

Run the interactive command line

Run the following to start interactive command line with default configuration

You can select a list of operation to perform

uv run main.py

The following table explain functionality of each option

Option	Description	Notes
1	Download the dataset
2	Extract & visualize labels
3	Compute FC
4	Reduce Feature Elimination	Not reported in the paper
5	Train and test classic models	Not reported in the paper
6	Train and test neural network models	Not reported in the paper
7	Train and test joint model
8	Test on new sites
9	Run ablation study
q	Quit

A recommended sequence of option to reproduce the results is as follows:

• (1): Download all labels and data
• (2): Extract cleaned label from the raw data
• (3): Choose 167 AAL and Compute FC for NYU dataset with 7 segmentation
• (7): Train and test the AECLS model (Reconstruction loss + Classification Loss)
• (8): Test AECLS model on new site (All other sites except NYU)
• (9): Train and test CLS model (No reconstruction loss)