#iid-assumption

Question 1: Select all that apply

Generalization guarantees assume that training and test examples are drawn IID from the same distribution. In which of these datasets are the examples dependent, so that a plain random train/test split would give a misleading test error? Select all that apply.

Several MRI scans per patient, split at random scan by scan1
Hourly temperature readings of one city, shuffled before splitting2
Survey answers from people sampled independently at random3
Successive rolls of a fair die, recorded in the order they happened4

Timed test