引導式測驗 · 中等 等級

選擇 K 的肘部法則測驗

檢查 k 均值慣性、說明性肘部、模糊曲線、縮放和初始化、間隙統計參考、輪廓和實用性。

相關引導路徑選擇 K 的肘部法則
問題 1 的 8

What does k-means inertia measure in the guide?

預覽每個問題8 問題

當您參加測驗時,就會出現答案和解釋。

  1. What does k-means inertia measure in the guide?
    • The share of future users who will join each cluster
    • The number of labels assigned incorrectly against ground truth
    • The distance between every pair of cluster centers only
    • The sum of squared distances from observations to their assigned centers
  2. Why does simply choosing the k with the lowest training inertia often overselect clusters?
    • More centers can keep reducing training distance without improving usefulness
    • Every added center proves a new natural category
    • Inertia necessarily rises as centers are added
    • The lowest inertia is calculated only for k = 1
  3. The guide's invented inertia scores for k = 1 through 5 are 120, 70, 35, 32 and 30. Which k is the elbow candidate?
    • One, before any reduction is observed
    • Five, because it has the smallest displayed inertia
    • Four, because 32 is closer to 30 than to 35
    • Three, where large reductions give way to small ones
  4. A real inertia curve bends gradually with no clear corner. How should the analyst describe the result?
    • Several k values may be defensible, so report ambiguity and use other checks
    • Select the lowest observed inertia despite the ambiguous bend
    • Treat one separation metric as conclusive without checking stability or purpose
    • Choose the first slight flattening as uniquely optimal without further checks
  5. Why can feature scaling change an elbow plot?
    • Scaling always makes every cluster perfectly round
    • The distance and inertia calculations depend on feature magnitudes
    • Scaling replaces the need to choose k
    • Only cluster names, not point distances, are affected
  6. What does the gap statistic compare with observed within-cluster dispersion?
    • The number of labels provided by a human annotator
    • The largest training inertia from a different dataset
    • Dispersion expected under an appropriate reference null distribution
    • The same curve after changing cluster names
  7. Which additional plot can help show whether candidate clusters are separated and whether some are weak?
    • Only the inertia curve, which measures within-cluster fit
    • A plot of within-cluster distances without comparison to neighboring clusters
    • A silhouette plot for the candidate k values
    • A table of cluster sizes without separation information
  8. Why should a team inspect cluster members and downstream use before treating an elbow as the right k?
    • Inertia already proves causal group membership
    • A geometric fit score does not establish meaningful or useful groups
    • An elbow makes labels inherently fair
    • Every elbow automatically transfers to new feature spaces