som-vae

Adaptation of the SOM-VAE code for fMRI generation purposes in Tensorflow v2.
Code adapted from https://github.com/ratschlab/SOM-VAE

As the initial code, scripts use the sacred librairy to monitor run parameters

Create training environment

To create the conda environment necessary to train the som-vae

conda env create -f environment.yml

Prepare tfrecords

To create tfrecords that will be used for training the som-vae. Data must be 2D Nifti files with shape [time_steps, sample_size]

python3 somvae_train.py with prepare=True tf_folder='my_folder' data_pattern='my_pattern'

Train som-vae with tfrecords

To train the som-vae with tfrecords using a certain batch size:

python3 somvae_train.py with prepare=False tf_folder='my_folder' batch_size=2