Solutions to the hands-on can be found on the corresponding page (clic on the "hands-on" menu on the left side).
For those of you who want to practice the concepts presented by Martin on some real data (to practice with software, not paper and pencil), we think that the famous "20 newsgroups" dataset is appropriate. It's available in scikit-learn, for instance.
There is also this nice blog page, which precisely illustrates evaluation on the "20 newsgroups" dataset and could serve as an inspiring starting point for some practice with this respect: https://krakensystems.co/blog/2018/text-classification.
