Ikibazo gihuza ibibazo · Biroroshye Urwego

Ikibazo cya Gini

Kubara umwanda wanduye hamwe no kwiyongera kuremereye, hanyuma utandukanye uruhare rwa Gini rwaho rwo kubaka ibiti no gusuzuma ibizaba.

Inzira zijyanye no kuyoboraGini Yanduye
Ikibazo 1 Bya 8

Ikibiri kinini gifite ibyiciro 0.8 na 0.2. Umwanda wacyo ni uwuhe?

Reba buri kibazo8 ibibazo

Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.

  1. Ikibiri kinini gifite ibyiciro 0.8 na 0.2. Umwanda wacyo ni uwuhe?
    • 0.32
    • 0.20
    • 0.64
    • 0.80
  2. Nuwuhe gaciro usobanura umwanda muri node irimo icyiciro kimwe gusa?
    • 0.5
    • 0
    • 1
    • It depends on the number of predictors
  3. Nigute umwanda wabana ugomba guhuzwa mugihe cyo gusuzuma abakandida batandukanijwe?
    • Add child scores without considering child sizes
    • Take the largest child impurity only
    • Weight each child by its share of the parent observations
    • Use only the purity of the smallest child
  4. Kugabanuka kwa Gini kugereranya niki mu kubaka ibiti?
    • The probability that a feature is causal
    • The fraction of training labels predicted incorrectly on test data
    • The number of classes in the target
    • Parent impurity minus weighted child impurity
  5. For two equally common classes in a binary node, what is Gini impurity?
    • 0
    • 0.5
    • 0.25
    • 1
  6. Ni irihe tandukaniro riri hagati ya Gini na entropiya?
    • Entropy is a distance between predictor vectors
    • Gini evaluates held-out accuracy while entropy builds trees
    • They are different local impurity criteria with different scales
    • Gini is always numerically larger for every split
  7. A deep tree has nearly pure training leaves. What should be checked before concluding it generalizes well?
    • Whether every feature has a positive value
    • Whether every leaf has zero entropy only
    • The alphabetical order of class labels
    • Performance on appropriately held-out data
  8. What does node-local Gini impurity describe?
    • Class mixing among observations in that node
    • The global class balance in every dataset
    • The probability the fitted model is calibrated
    • Feature importance across all possible models