Quiz collegato alla guida · Difficile Livello
LIME Spiegazioni locali Quiz
Comprendere il surrogato locale di LIME, le scelte di perturbazione e ponderazione, i controlli di stabilità e i limiti dell'interpretazione dei contributi delle funzionalità.
Domanda 1 di 8
Cosa si adatta LIME alla previsione selezionata?
Anteprima di ogni domanda8 domande
Le risposte e le spiegazioni vengono visualizzate mentre rispondi al quiz.
- Cosa si adatta LIME alla previsione selezionata?
- A replacement model guaranteed to match globally
- A causal model of the observed real-world outcome
- A simple interpretable surrogate approximating the original model locally
- A single global feature ranking shared by all examples
- Perché le perturbazioni vicine hanno un peso maggiore in una spiegazione locale?
- To make every perturbed feature equally common
- To focus the surrogate on behavior near the instance
- To guarantee perturbations match the true data distribution
- To give distant decision regions equal influence on the local fit
- Una spiegazione LIME cambia drasticamente quando cambia il seme del campionamento. Cosa dovrebbe fare l’analista?
- Report only the explanation with the clearest feature signs
- Assume the original predictor's outputs must also change between runs
- Treat sampling variation as proof of a causal relationship
- Assess stability under reasonable settings before relying on it
- Perché la rimozione delle parole in modo indipendente può creare un quartiere testuale inadeguato?
- The resulting sentences may be unnatural
- The original word order is guaranteed to remain semantically equivalent
- Independent removals guarantee samples from the true language distribution
- Local weighting automatically restores missing grammatical relationships
- Cosa stabilisce il contributo di una funzione LIME?
- The real-world cause of the outcome
- Global feature importance for every example
- A local association in a surrogate explanation, not causation
- That a prediction is correct
- Which diagnostic directly checks whether a local surrogate represents the original model near an instance?
- Measure the predictor's global training accuracy alone
- Measure surrogate fidelity in the weighted neighborhood
- Count only the number of selected surrogate features
- Inspect only the surrogate coefficient signs
- For an image explanation, what can change which regions appear important?
- Only the display order of identical highlighted regions
- Only the text used to name an unchanged region
- Only the image file's name when pixels stay fixed
- The segmentation used as interpretable units
- Which conclusion is appropriately limited after inspecting one LIME explanation?
- This surrogate approximates this model near this selected input under stated settings
- The model uses features identically for all cases
- The highlighted factor caused the real outcome
- The whole model is fair and trustworthy