bad labels: Bad labels are a genuine concern. When you make a model better by tuning the hyperparameters, how can you be sure you're not overfitting on bad labels? Odds are that efforts doing GridSearch are better spent checking for bad labels, especially if you haven't done so already.
So please go out there, any find some bad labels!
Notebook
If you're interested in the entire notebook from this series, you can find it on Github.