Kvíz spojený s průvodcem · Střední úroveň
Kvíz izolace lesa
Propojte náhodné délky cest stromů s hodnocením anomálií a odlište práh kontaminace od skutečné míry anomálie.
Otázka 1 z 8
Co se obvykle stane s průměrnou délkou cesty izolačního stromu neobvyklého bodu?
Náhled každé otázky8 otázky
Odpovědi a vysvětlení se zobrazují, jak budete v kvízu.
- Co se obvykle stane s průměrnou délkou cesty izolačního stromu neobvyklého bodu?
- It is shorter because fewer random splits isolate it
- It is longer because every split keeps it with common points
- It equals the number of features exactly
- It is set by contamination directly
- Co běžně kontroluje nastavení kontaminace?
- The verified prevalence of anomalies in production
- The threshold used to label a chosen fraction as outliers
- The number of input features
- The random split direction for every tree
- Která interpretace skóre odpovídá základní intuici lesa?
- Longer paths always mean higher anomaly risk
- A score is a calibrated fraud probability
- Shorter paths indicate stronger isolation-based anomaly evidence
- The score identifies the causal reason for an event
- Vzácný, ale legitimní segment zákazníků získává vysoké skóre anomálií. Jaký problém to ilustruje?
- Path lengths are always causal
- Contamination guarantees correct labels
- Random trees require class labels
- Rarity is not equivalent to fraud or invalidity
- Proč ověřovat směr skóre ve zvolené knihovně?
- All libraries define negative scores as normal
- Some decision functions invert or shift the raw anomaly score
- The tree ensemble omits path lengths
- Contamination reverses the feature matrix
- If contamination increases from 5% to 10%, what change is most directly expected?
- Every path becomes shorter
- The true anomaly rate is proven to double
- More points may be labeled anomalous by the threshold
- The model changes into supervised fraud classification
- How should an alerting team evaluate an Isolation Forest threshold when labels are available?
- Choose the threshold that labels exactly half the data
- Assume every short path is fraud
- Compare only the training score average
- Measure precision and recall tradeoffs on representative labeled cases
- What can an Isolation Forest score not provide by itself?
- A causal explanation or calibrated event probability
- A ranking of isolation behavior
- An ensemble-based anomaly score
- A comparison across observations