引導式測驗 · 中等 等級
Silhouette 分數和聚類評估測驗
計算每點輪廓值並將內部幾何診斷與群集效用和替代指數區分開來。
問題 1 的 8
當a=2且b=5時,輪廓係數(b-a)/max(a,b)是多少?
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。
- 當a=2且b=5時,輪廓係數(b-a)/max(a,b)是多少?
- 0.6
- -0.6
- 0.4
- 1.5
- 一個點有 a=4 和 b=3。它的負輪廓顯示什麼?
- Its assigned cluster has zero observations
- It is, on average, closer to another cluster under the chosen distance
- The global score must be negative
- The feature scaling is proven correct
- 對於一次觀察,接近零的輪廓通常顯示什麼?
- Its class label is known
- It is perfectly separated from all clusters
- It lies near a boundary between its own and another cluster
- The clustering has one cluster only
- Davies-Bouldin 指數通常首選哪個方向?
- A value matching the silhouette range
- Higher values
- Exactly zero for every useful partition
- Lower values
- 為什麼原始 Davies-Bouldin 值和 Calinski-Harabasz 值不應該直接比較?
- Both are probability estimates between zero and one
- They use different formulas, scales and preferred directions
- They require identical units by definition
- One is always the reciprocal of the other
- Why can an average silhouette hide a cluster-specific problem?
- Every sample receives a different formula
- Silhouette uses only labels and no distances
- A sample-weighted mean can mask weak values in a smaller group
- A negative value is omitted automatically
- What does a high silhouette score fail to establish by itself?
- That labels were assigned
- That within and between distances were calculated
- That the chosen metric was used
- That the clusters are useful for the downstream task
- When is the silhouette score undefined in common implementations?
- For degenerate labelings with one cluster or one cluster per sample
- Whenever the average coefficient is positive
- When a cluster has two observations
- When distances use standardized features