引導式測驗 · 中等 等級

核心密度估計測驗

測試 KDE 內核和頻寬選擇、特徵縮放以及平滑如何影響密度估計的解釋和驗證。

相關引導路徑核心密度估計
問題 1 的 8

KDE 從觀察樣本估計出什麼?

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. KDE 從觀察樣本估計出什麼?
    • A classification boundary with guaranteed accuracy
    • A smooth estimate of a continuous probability density
    • A list of exact population probabilities for every value
    • A causal effect of one feature on another
  2. 頻寬如何影響 KDE 平滑?
    • It sets the number of labels in the training set
    • It determines the true number of populations
    • It is the probability assigned to each sample point
    • It controls the width of smoothing around each data point
  3. What can happen when bandwidth is too small?
    • Random sample variation may appear as many narrow peaks
    • Distinct modes are always merged into one flat curve
    • The estimate becomes a histogram with fixed bin labels
    • Every density value becomes a calibrated posterior probability
  4. What may happen when bandwidth is too large?
    • Kernel choice no longer affects any density estimate
    • The sample observations are removed from the calculation
    • Separate clusters or tails can be flattened or merged
    • The estimate becomes more jagged around every observation
  5. Why should features be scaled before multivariate KDE in some datasets?
    • Different units can make distance-based kernels dominated by one large-scale feature
    • Scaling guarantees the bandwidth is optimal
    • KDE requires all variables to be binary
    • Scaling removes missing data automatically
  6. 交叉驗證有助於在 KDE 工作流程中選擇什麼?
    • The true number of modes in every population
    • The measurement units for each feature
    • A bandwidth or model setting under a specified density objective
    • Which observations should be considered causal
  7. 為什麼要檢查 KDE 圖的幾個合理頻寬?
    • To ensure every curve has the same number of peaks
    • To select the kernel with the largest density values
    • To replace the need to compare with the raw sample
    • To see whether apparent features persist or are artifacts of a smoothing choice
  8. How should a KDE curve be interpreted for a continuous variable?
    • As a histogram count with no dependence on bandwidth
    • As estimated probability per unit of measurement, not probability at one exact value
    • As the probability that a point equals a particular real number
    • As a guaranteed description of the population distribution