Clustering: Structure Without Labels · seed 1 · A4, ink-friendly. The answer key prints on its own page for grown-ups.

Groups with no right answers attached

Computing · Machine Learning · ages 20-22
Name ______________________   Date ____________

  1. What must you decide before running k-means?

    • The number of centres k and the distance to use
    • The correct labels for every single point in advance
    • The file name of the dataset

  2. What two steps does k-means alternate?

    • Assign points to the nearest centre, then move each centre to its group middle
    • Delete the farthest point, then grow a new centre from all the remaining points
    • Sort labels alphabetically, then count them

  3. Clustering starts with no target column, asking only which examples resemble each other.

    Circle one:   True   False

  4. Your data forms long stretched groups with stray points. Which tool fits better?

    • K-means, since it loves stretched shapes
    • DBScan, which walks neighbours and can label strays as noise
    • No algorithm fits stretched data, so grouping is hopeless here

  5. Two runs on the same points end with different groups. What best explains it?

    • The points moved between runs
    • K-means always returns the same groups on every run from any start
    • Different starting centres settled into different final groups

  6. What makes a clustering result meaningful rather than merely tidy?

    • Groups are tight inside, well separated, and say something true
    • The algorithm finished without errors, which proves the groups real
    • The groups look neat in a table

  7. A friend says the groups must be real because the algorithm finished. What is the best reply?

    • True, finished output is always meaningful
    • K-means can force round groups that are not real
    • Rerun from fresh starts until the groups look rounder

  8. Points sit at 2, 4, 10 and 12, with starting centres at 2 and 12. After one round of assigning points and moving centres, where does the first centre sit?

    Answer: ______________

LightMySky · lightmysky.comW1-mt_BAMGOlWgRY-s1

Answer key

For grown-ups. Fold this page away before handing over the rest.

Groups with no right answers attached W1-mt_BAMGOlWgRY-s1

  1. The number of centres k and the distance to use · K-means cannot start without k and a way to measure nearness.
  2. Assign points to the nearest centre, then move each centre to its group middle · Assign then move, repeated until the groups stop shifting.
  3. True · No labels are given, so resemblance is the only guide.
  4. DBScan, which walks neighbours and can label strays as noise · K-means builds round blobs around centres, while neighbour-walking follows stretched shapes.
  5. Different starting centres settled into different final groups · The starting centres steer where assigning and moving converge.
  6. Groups are tight inside, well separated, and say something true · Finishing proves nothing. Tight, separated and true groups are what prove the result.
  7. K-means can force round groups that are not real · Finishing only means the steps ran. A wrong k can tidy noise into false blobs.
  8. 3 · 2 and 4 are nearer to the centre at 2, so the first centre moves to their middle, which is 3. The second centre takes 10 and 12 and moves to 11.
Worksheet · LightMySky