Dynamic Time Warping Quiz

Answers and explanations appear as you take the quiz.

1. What variation is DTW designed to accommodate when comparing sequences?
   • Different timing or execution speed
   • Different class-label names
   • Arbitrary changes in feature meaning
   • Missing all observations
2. What does the dynamic-programming cost matrix accumulate?
   • The average value of both sequences
   • The lowest allowed path cost up to each pair of positions
   • The number of labels per sequence
   • A probability that the alignment is causal
3. Why constrain the warping path near the diagonal?
   • To make all distances satisfy triangle inequality
   • To force every pair to match exactly at the same index
   • To prevent implausible extreme timing matches
   • To remove local costs from the calculation
4. What can unrestricted warping do to unrelated sequences?
   • Preserve sampling rates automatically
   • Guarantee a semantic match
   • Make every local cost zero
   • Align distant or mismatched events through excessive stretching
5. Which quantity commonly defines the cost in one cell of the DTW matrix?
   • The product of sequence lengths
   • A distance between feature vectors at two time positions
   • The cluster count
   • The global mean of each series only
6. Why should normalization be consistent across sequences?
   • A DTW path has no cumulative cost
   • Normalization forces all alignments to be identical
   • Path-length normalization changes the scale of reported distances
   • The method becomes a classifier automatically
7. Which property is not guaranteed for standard DTW distance?
   • Matching both endpoints
   • A monotone alignment path
   • A cumulative local cost
   • Triangle inequality
8. What should a reviewer inspect alongside a DTW score?
   • The alignment path and whether its timing distortions are plausible
   • Only the sequence length
   • The alphabetical order of frames
   • The number of clusters in an unrelated model