This repository is the official implementation of the paper https://arxiv.org/abs/2101.04117, containing training code, evaluation scripts, and figure generation.
For an easy-to-use API implementing a trained version of this model, check out spock!
To train the model, download the data into the /data folder
from this Globus link.
Run the script train.sh, which will train
the model 30 times from different seeds.
To generate the figures, edit the figures/generate.sh script
to have true instead of false for any figure you'd like to generate.
Then, execute that script in the folder.
One can install a working conda environment using create_env.sh.
This file also contains a list of packages required by this model.