Double descent experiments

This is messy, but near the bottom of the notebook I managed to get some double descent, when 30% label noise was introduced.