Change Point Detection Quiz

Answers and explanations appear as you take the quiz.

1. Which kinds of series behavior can a change point mark?
   • A shift in mean, variance, trend or distribution
   • Only a change in class labels
   • Only the first observation
   • A guaranteed causal intervention
2. In a penalized offline segmentation objective, what does a larger penalty generally do?
   • Force a change at every observation
   • Discourage adding extra segments
   • Remove all segment costs
   • Increase the number of model variables
3. What does CUSUM accumulate over time?
   • Only the single largest observation in the sequence
   • The number of free model parameters
   • Evidence of deviation from a reference process
   • The moving average of observations not yet collected
4. Which setting is especially important for CUSUM alert behavior?
   • The class encoding order
   • The number of clusters
   • The metric denominator
   • The control threshold
5. What does Bayesian online change-point detection update?
   • A single offline segmentation with no uncertainty
   • A probability distribution over run lengths since a change
   • The target's class labels
   • The model's seasonal period only
6. Why is a detected change not itself an explanation?
   • The algorithm always outputs a causal variable
   • Change points never correspond to real events
   • A statistical shift may come from process change, seasonality, drift or data issues
   • A segmentation has no time index
7. What risk can a very low segmentation penalty create?
   • Guaranteeing fewer false alarms
   • Making every series stationary
   • Removing autocorrelation by construction
   • Over-segmenting random noise
8. Which evaluation concern matters for repeated online monitoring?
   • False alarms and detection delay across repeated looks
   • Only one p-value at the final time
   • Whether all segment means are identical
   • The maximum number of predictors