Appearance
Topic Drill 6.1.5: Metamorphic Testing (K3)
About this drill
LO AI-6.1.5 (K3): Use metamorphic testing to derive test cases for a given scenario.
Lesson: Ch 6, section 6.1.5, including the worked example, MR template and exam technique
This drill covers every point in section 6.1.5:
- Source and follow-up test cases; metamorphic relations based on required properties
- Consistency, monotonicity, invariance; functional and non-functional use; application areas
- When MT is chosen over oracle-based testing; the age-at-death example
- Deriving and validating MRs; limitations (incorrect or incomplete MRs, absolute errors); combining MT with other techniques
Topic Drill: 6.1.5 (K3)
Q1. An image classifier labels a source photo (which passed) as "cat". MR: "rotating the image by 5° must not change the class" (invariance). Which follow-up result REVEALS a failure?
ARotated photo → "cat" with 97% confidence
BRotated photo → "dog"
CRotated photo → "cat" with 92% confidence
DOriginal photo again → "cat"
Q2. A model predicts a house's daily heating energy. Source test case (passed): outside temperature 5 °C → 40 kWh. MR: "a lower outside temperature, all else equal, never needs LESS heating energy". Which follow-up result REVEALS a failure?
A0 °C → 46 kWh
B−5 °C → 52 kWh
C5 °C → 40 kWh
D−2 °C → 37 kWh
Q3. A hotel search returns {H1, H2, H3, H4} for the source query "breakfast: don't care" (passed). Follow-up T1: "breakfast: included". Follow-up T2: "breakfast: not included". All other criteria are unchanged. Which pair of follow-up results is MOST likely valid?
AT1: {H2, H4} · T2: {H1, H3}
BT1: {H2, H4, H5} · T2: {H1, H3}
CT1: {H2, H4} · T2: {H1, H2, H3}
DT1: {H2} · T2: {H1, H3}
Q4. Which metamorphic testing objective is defined as "outputs remain stable under perturbations"?
AConsistency
BMonotonicity
CInvariance
DTransferability
Q5. A route optimizer is tested with the MR "closing a road on the current best route never produces a SHORTER best route". Which objective does this MR check?
AInvariance
BMonotonicity
CConsistency
DLabel correctness
Q6. Where do testers derive metamorphic relations from, according to the syllabus?
AFrom domain knowledge, requirements, or domain properties such as the laws of physics
BOnly from the model's source code
CFrom the training data labels
DFrom the results of failed test cases only
Q7. How can metamorphic relations themselves be validated?
AThey cannot be validated
BBy running the model in production
CBy measuring neuron coverage
DBy expert review, by running them on reference models, and by checking coverage of edge cases
Q8. All metamorphic relations for a price-prediction model pass, but every prediction is in fact 10% too high. What does this illustrate?
AThe MRs must be incorrect
BThe model is underfitted
CMetamorphic testing detects relational flaws but not all absolute errors, so it should be combined with other test techniques
DMetamorphic testing cannot be used for regression models
Q9. The MR "higher income → never a lower credit limit" fails for some applicants who also have very high debts. What should the tester consider FIRST?
AReporting a model defect immediately
BWhether the MR is incorrect because it overlooks a complex interaction between variables (here, debt), and having it reviewed by a domain expert
CDeleting those applicants from the test set
DSwitching to A/B testing
Q10. Why is a metamorphic test usually based on a source test case that PASSED?
ABecause failed tests cannot be re-run
BBecause only passed tests have inputs
CBecause the tool requires it
DBecause MT checks relationships between outputs; if the source output were already wrong, the follow-up could satisfy the relation while also being wrong
Q11. Which TWO statements about metamorphic testing are CORRECT?
Select 2 options.
AMT requires an exact expected result for every follow-up test case
BMT can only be used in white-box testing
CMT can be applied to both functional and non-functional testing
DA single MR is always sufficient for confidence
EMT has been used for testing image recognition, search engines, route optimization and voice recognition
0 / 11 answered
← 6.1.4 Adversarial Testing · Next drill: 6.1.7 Drift Testing →