引導式測驗 · 硬 等級
目標編碼測驗
使用類別平均值、稀有倉庫範例和未見過的類別處理檢查目標編碼洩漏、交叉擬合和平滑。
問題 1 的 8
A category occurs in one training row. Why is using that row's unsmoothed target mean as its own input dangerous?
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。
- A category occurs in one training row. Why is using that row's unsmoothed target mean as its own input dangerous?
- The encoding necessarily becomes identical to the global mean
- The category frequency alone reveals the model's validation score
- The encoding directly reveals that row's outcome
- An unsmoothed single-row encoding is protected from leakage by its small sample size
- How should a held-out fold receive target encodings during cross-fitting?
- From category summaries calculated using the other training folds
- From summaries calculated using its own labels only
- From summaries calculated using the final test labels
- From summaries calculated using all training folds, including its own labels
- Using the guide's smoothing example, what value results from a target sum of 30, count of two, global mean six and prior weight four?
- 15
- 6
- 54
- 9
- Why does smoothing pull a rare category's estimate toward the global training mean?
- To make every category's encoding identical
- To reduce the influence of very small category samples
- To remove the need for an evaluation split
- To guarantee that category frequency cannot change
- Which scikit-learn TargetEncoder operation uses internal cross-fitting when encoding training data?
- Calling transform on training rows using a mapping fitted on all those rows
- Calling fit and then transform on the same training rows
- Calling fit_transform on the training data
- Calling fit once before dividing the training data into evaluation folds
- How should evaluation labels be used when constructing a target-encoding mapping?
- They should not be used to learn the mapping being evaluated
- They should define the mapping for rare categories only
- They should determine the global mean but not category means
- They should be added when the training score is low
- According to the guide, what does scikit-learn's TargetEncoder use for a category absent from its fitting data?
- The smallest observed category value
- The held-out outcome for that row
- A newly calculated test-set average
- The learned target mean
- A target-encoding dataset contains repeated entities and time-dependent outcomes. Which design needs attention beyond ordinary random cross-fitting?
- Using random row folds because each row excludes its own label
- Choosing splits that respect the intended group and time boundaries
- Using stronger smoothing as a substitute for respecting time boundaries
- Using an ordinary Pipeline as proof that its encoder's internal folds respect groups