引導式測驗 · 中等 等級
機器學習面試問題小測驗
測驗評估、正規化、衡量權衡,以及公共面試指導如何在不承諾提供確切問題清單的情況下為準備提供資訊。
問題 1 的 8
Why can accuracy be misleading on a highly imbalanced dataset?
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。
- Why can accuracy be misleading on a highly imbalanced dataset?
- Accuracy is independent of the class distribution
- Accuracy always equals recall
- A majority-class predictor can score well while missing most rare positives
- Accuracy ignores every correctly predicted negative
- Which metric directly measures the fraction of predicted positives that are correct?
- 精確度
- 召回
- 準確度
- False-negative rate
- How should a candidate choose between precision and recall priorities?
- Always maximize precision regardless of the application
- Choose the metric with the largest numeric value
- Ignore the prediction threshold
- Relate the choice to the relative cost of false positives and false negatives
- What does regularization seek to balance during model training?
- Replace validation with training performance
- Fit to the data while limiting unnecessary model complexity
- Increase training error without considering generalization
- Maximize the number of features regardless of evidence
- A model performs much better on training data than validation data. Which next step is most grounded?
- Add features before checking the split
- Change the test labels to match training
- Check generalization, data splits, and possible leakage before choosing a fix
- Report the training metric as expected production quality
- Which preparation areas does Microsoft’s public guide name for AI/ML interviews?
- Only memorizing product names
- Only writing research papers
- Only designing mobile interfaces
- Model evaluation, profiling, optimization, and applying ML principles
- How should candidates treat example interview prompts in this guide?
- As a certification exam
- As practice prompts, not a guaranteed list of questions
- As the exact questions used by every employer
- As a substitute for the job posting
- How can a candidate make a model-tradeoff answer useful?
- State assumptions, baseline, metric, and what evidence could change the decision
- Name the most complex model available
- Avoid discussing data or errors
- State a metric without explaining its relevance