Kvíz spojený s průvodcem · Střední úroveň
Pravděpodobnost a statistika pro ML Kariérní kvíz
Použijte vzorkování, průměry, podmíněné pravděpodobnosti a nejistotu k interpretaci důkazů ML, včetně přesnosti, zapamatování a závislých pozorování.
Otázka 1 z 8
For [0, 0, 0, 20], which mean and median are correct?
Náhled každé otázky8 otázky
Odpovědi a vysvětlení se zobrazují, jak budete v kvízu.
- For [0, 0, 0, 20], which mean and median are correct?
- Mean 5 and median 0
- Mean 0 and median 5
- Mean 20 and median 0
- Mean 5 and median 10
- A system issues 100 alerts, of which 10 identify true events. What is its precision?
- 100 %
- 10 %
- 1%
- 90 %
- The same system flags all 10 true events present in the dataset. What is its recall?
- 10 %
- 90 %
- 100 %
- 1%
- A dataset adds many more rows collected through the same biased selection process. What is not guaranteed?
- That the row count increased
- That the dataset occupies more entries
- That a sample mean can be calculated
- That the sample now represents the intended population
- Why might repeated records from one person need to stay together in a holdout split?
- They can share information, so splitting them across training and testing can weaken independence
- Every record from a person is necessarily identical
- Grouping guarantees the model will generalize
- A holdout set must contain exactly one row
- Under its assumptions, what does a frequentist 95% confidence procedure describe?
- A guarantee for every individual interval
- Its long-run coverage of the fixed parameter across repeated samples
- The probability that every observed data point is correct
- A promise that the model has 95% classification accuracy
- An ML study reports a correlation between an input and an outcome. Which inference requires more than that association?
- The variables are associated in the observed analysis
- The study computed a relationship
- Changing the input would cause the predicted outcome change
- A sample was used in the analysis
- Before comparing precision across two model runs, what should be checked?
- Only that both percentages have one decimal place
- Only that the models share an architecture
- Only that the test files have equal size
- The labels, threshold, evaluation period and denominators