Multimodal Parkinson's

TODO

There's a few #Todo's in the code, need to take a look at them

Verify quality of latent vectors by clustering into healthy and PD patients

Incorporate more modalities and ensure data_init and data_info can track multiple images per entry

Convert to beta VAE learning Bernoulli?

Implement Spatial Broadcast Decoding instead of conv transpose

Running Code

All commands should be run from the root github directory. Run python src/main.py --help for more information about arguments.

Training

python src/main.py --train [ARGS] where args can be -b BATCH_SIZE -e EPOCHS --save_frequency FREQUENCY. It is recommended to train the network with nohup python src/main.py --train [ARGS] & to avoid stopping training due to SSH disconnecting or any other reason.