We need to make sure that running `gaussian.py` (possibly on a smaller data set than in the original paper) still gives results that make sense.