引導式測驗 · 中等 等級

機器學習中的距離度量測驗

測試歐幾里德距離、曼哈頓距離、餘弦距離、馬哈拉諾比斯距離和漢明距離如何編碼相似性,以及為什麼特徵縮放可以改變最近鄰結果。

問題 1 的 8

為什麼配送路徑規劃示例選擇曼哈頓距離而非歐幾里得距離?

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. 為什麼配送路徑規劃示例選擇曼哈頓距離而非歐幾里得距離?
    • Because vehicles must travel along street grids rather than in straight lines
    • Because Manhattan distance is always more accurate
    • Because Manhattan distance requires less computation for any dataset
    • Because Euclidean distance cannot handle numeric data
  2. 根據指南定義,餘弦距離測量什麼特性,使其適合推薦系統?
    • The absolute difference in rating counts
    • The number of matching categorical labels
    • The angle between two vectors rather than their magnitude
    • The square of the Euclidean distance
  3. 馬哈拉諾比斯距離比歐幾里得距離多納入了什麼額外資訊?
    • The alphabetical order of feature names
    • The correlations and differing variances between features via the covariance matrix
    • The total number of rows in the dataset
    • The categorical labels of each point
  4. 根據指南,漢明距離特別適合什麼類型的資料?
    • Continuous numeric data with large ranges
    • Only image pixel data
    • Only time-series financial data
    • Equal-length categorical or string data, such as genetic sequences
  5. 該指南說在計算跨特徵的歐幾里德距離之前通常需要什麼預處理步驟?
    • Removing all categorical variables
    • Standardizing or scaling the features so no single feature dominates
    • Converting all features to Hamming-compatible strings
    • Reducing the dataset to two dimensions only
  6. What phenomenon does the guide describe that undermines the assumption that Euclidean distance is always a safe default?
    • 過擬合
    • Gradient vanishing
    • The curse of dimensionality, where pairwise distances become similar in high dimensions
    • 標籤洩漏
  7. How is cosine distance computed, per the technical section?
    • As one minus the cosine similarity, the normalized dot product of two vectors
    • As the sum of absolute differences per dimension
    • As the square root of the covariance matrix
    • As the count of mismatched positions between two vectors
  8. According to the futureOutlook section, how does metric learning relate to distance metrics in modern deep learning systems?
    • It eliminates the need for any distance function
    • It only applies to Manhattan distance
    • It replaces Hamming distance in all genetic applications
    • A neural network learns an embedding space where a classical distance, like cosine or Euclidean, becomes meaningful after training