引導式測驗 · 簡單 等級

機器學習基礎知識測驗

對實際機器學習任務進行分類、比較基準、保護測試資料並發現平均值可能隱藏的泛化或部署故障。

相關引導路徑機器學習基礎知識
問題 1 的 8

A store wants to estimate next week’s sales as a dollar amount from past sales and calendar features. Which task is this?

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. A store wants to estimate next week’s sales as a dollar amount from past sales and calendar features. Which task is this?
    • 回歸
    • 分類
    • 聚類
    • Text generation
  2. A support team assigns each incoming message to one of five known issue labels. Which task matches that goal?
    • 聚類
    • 分類
    • 回歸
    • Dimensionality reduction
  3. An analyst groups unlabeled news articles by similarity without predefined topics. Which task is being used?
    • 分類
    • 回歸
    • 聚類
    • Forecasting
  4. A model scores 98% on training records but performs poorly on new records. What is the main concern?
    • It has too few input columns by definition
    • It has proved that the labels are correct
    • It is necessarily a clustering model
    • It may fit the training examples well without generalizing to new inputs
  5. Why should a team keep a final test set out of repeated model tuning?
    • Repeated tuning can adapt decisions to the test set, weakening it as an independent check
    • The test set is only for fitting model parameters
    • A test set prevents selecting a metric
    • The training set must always be smaller
  6. A forecasting team compares a complex model with “use yesterday’s value” before adopting it. What purpose does that simple rule serve?
    • A substitute for monitoring after deployment
    • A baseline showing whether added complexity provides useful information
    • A way to relabel regression as classification
    • A guarantee that the complex model is fair
  7. A deployed model suddenly receives temperatures in Fahrenheit after training on Celsius. Which workflow lesson applies?
    • Strong test accuracy makes input units irrelevant
    • Only the model architecture can cause failures
    • Errors can arise from production preparation or changed input conventions
    • Deployment removes the need to check inputs
  8. A classifier has strong average accuracy but misses nearly every rare emergency case. What should the team do?
    • Accept it because average accuracy is high
    • Tune on the final test set until the score improves
    • Remove the rare cases from evaluation
    • Inspect failure cases and choose metrics that reflect the consequences of mistakes