Databricks Machine Learning Professional Question 256
Select 2You are tasked with monitoring numerical data for drift in an e-commerce platform's user behavior features. You are considering using either the Jensen-Shannon (JS) divergence or the Kolmogorov-Smirnov (KS) test. Which of the following statements correctly compare these methods for numerical drift detection?
- A
The Kolmogorov-Smirnov test is non-parametric, while the Jensen-Shannon divergence assumes a probabilistic distribution for comparison.
- B
Jensen-Shannon divergence provides a symmetric measure of how two probability distributions differ, while the Kolmogorov-Smirnov test focuses on the maximum difference in cumulative distributions.
- C
The Kolmogorov-Smirnov test requires binning of numerical data before application, while Jensen-Shannon divergence does not.
- D
Jensen-Shannon divergence outputs a specific statistical p-value to determine significance, while the Kolmogorov-Smirnov test does not.
- E
The Kolmogorov-Smirnov test is more suitable for detecting changes in feature distributions over time, while Jensen-Shannon divergence is better for comparing static distributions.
Show answer and explanation
Correct answers: A, B
Explanation
The Kolmogorov-Smirnov (KS) test and Jensen-Shannon (JS) divergence are both used for drift detection, but they differ in approach. The KS test is non-parametric and identifies the maximum difference in cumulative distributions, making it useful for hypothesis testing with a p-value output. JS divergence measures dissimilarity in a symmetric way, assuming probability distributions as input. These distinctions make them suitable for different scenarios depending on the data type and analysis goals.
- A. Correct.
Correct. The Kolmogorov-Smirnov test is non-parametric and directly compares cumulative distributions, while JS divergence operates on probability distributions and assumes they are well-defined.
- B. Correct.
Correct. JS divergence measures the dissimilarity between two distributions in a symmetric manner, whereas the KS test identifies the largest deviation between cumulative distributions.
- C. Incorrect.
Incorrect. The Kolmogorov-Smirnov test does not require binning of numerical data, as it directly measures differences in cumulative distributions.
- D. Incorrect.
Incorrect. Jensen-Shannon divergence does not output a p-value; it simply provides a divergence score, while the KS test does output a p-value to indicate statistical significance.
- E. Incorrect.
Incorrect. Both the Kolmogorov-Smirnov test and Jensen-Shannon divergence can be used for static or time-series comparisons, but the choice depends on the type of drift and the data characteristics.