Experiment to determine whetever a large batch-size can be helpful with extremely umbalanced datasets.
As usual, just download it using pip:
pip install udbnn
Since some software handling coverages sometime get slightly different results, here's three of them:
Since the experiments take quite a bit to run, I suggest you to run them while in a TMUX-like environment. If available, you should consider using a computer with a tensorflow-compatible GPU.
Then just run with a python shell:
from udbnn import run
run("dataset")