引導式測驗 · 中等 等級
F1 分數與 F‑Beta 測驗
測試調和 F1 公式、基於計數的範例、F-beta 權重、宏平均與微平均、閾值效應以及單一分數的極限。
問題 1 的 8
F1 如何結合精確度與召回率?
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。
- F1 如何結合精確度與召回率?
- 作為兩個值中較大的那個
- 作為它們的調和平均數
- 作為召回率減去精確度
- 作為所有預測中為負的比例
- 在指南構造的 8 TP、2 FP 和 8 FN 範例中,F1 約為多少?
- 0.615,來自 16/(16+2+8)
- 0.80,等於精確度
- 0.50,等於召回率
- 0.90,僅來自正確的正向預測
- 當 F-beta 中的 beta 大於一時,哪個組件獲得更高權重?
- 精確度,因為假陽性成為唯一錯誤
- 整體準確度,無論正類如何
- Specificity, replacing precision and recall
- Recall, while precision still contributes
- A task values avoiding false alerts more than catching every positive case. Which beta direction can reflect that priority?
- A beta greater than one to emphasize recall
- A positive beta below one to emphasize precision
- Beta exactly one because it is always optimal
- A negative beta to reverse the F-score
- What does macro F1 do across multiple classes?
- Compute each class's F1 and average the class scores equally
- Pool all TP, FP and FN before calculating one score
- Weight each class solely by its prevalence
- Use only the largest class's F1
- How is micro F1 formed for a multiclass report?
- Average per-class F1 without weights
- Take the median of per-class F1 scores
- Use only the rarest class
- Aggregate TP, FP and FN across classes before computing F1
- A model's predicted probabilities are unchanged, but its classification threshold moves. What can happen to F1?
- It must stay fixed because model weights did not change
- It becomes a direct probability-calibration score
- It can change as precision and recall trade off
- It changes only when the test-set size changes
- Which concern is not established by a strong overall F1 score alone?
- That the formula combined precision and recall
- That a threshold and positive label were used
- 機率經過校準或不存在子群組傷害
- 可以報告真陽性和偽陽性的計數