Exam AWS Certified Machine Learning - Specialty topic 1 question 227 discussion

Jensen-Shannon divergence, Kullback-Leibler divergence, and Total variation distance—are all measures of statistical distance between probability distributions. They are useful for understanding how different two distributions are, but they are not specifically designed to measure bias in categorical variables.

It is leaning towards bias in data, rather than probability distribution.

BIAS in _data_ before training uses following matrices B - helps assess whether there is bias in how loan amounts are distributed among different categories C - ompares the proportions of positive (e.g., approved loans) and negative (e.g., rejected loans) outcomes across different facets (demographic groups) F - High total variation distance between between the predicted and observed labels suggests potential bias

Question asks for distributions. DEF are distributions. ABC are imbalances or disparities.

Since it is indicated in the official web site that D, E and F are used to determine how different the distributions for loan application outcomes are for different demographic groups https://docs.aws.amazon.com/sagemaker/latest/dg/clarify-measure-data-bias.html

All valid answers ? They are listed as "pre-training bias" here https://docs.aws.amazon.com/sagemaker/latest/dg/clarify-measure-data-bias.html

confusing but lean towards B D and F

All are valid answers, so definitely un scored question https://docs.aws.amazon.com/sagemaker/latest/dg/clarify-measure-data-bias.html

answers "How different are the distributions for loan application outcomes for different demographic groups?"

https://docs.aws.amazon.com/sagemaker/latest/dg/clarify-measure-data-bias.html

D. Jensen-Shannon divergence and E. Kullback-Leibler divergence are post-training bias metrics that measure the distance between two probability distributions. They are not pretraining bias metrics and cannot be used to check the bias distribution of the dataset. F. Total variation distance is a post-training bias metric that measures the difference between two probability distributions. It is not a pretraining bias metric and cannot be used to check the bias distribution of the dataset. Send a message...

Agree with austinoy, answer should be DEF.

https://docs.aws.amazon.com/sagemaker/latest/dg/clarify-measure-data-bias.html