jmalkin commented on code in PR #211: URL: https://github.com/apache/datasketches-website/pull/211#discussion_r1816976465
########## docs/QuantilesAll/QuantilesOverview.md: ########## @@ -21,92 +21,99 @@ layout: doc_page --> # Introduction to the Quantile Sketches -This is a quick overview of the quantiles sketches in the library. Each of these quantile types may have one or more specific implementaions and some variation in API depending on the language. Three of the quantile sketches have mathematically provable error bounds while the fourth is an empirical algorithm. +This is an overview of the quantiles sketches in the library. Each of these quantile types may have one or more specific implementaions and some variation in API depending on the language. Three of the quantile sketches have mathematically provable error bounds while the fourth is an empirical algorithm. The three sketches with mathematically provable error bounds are: * The Classic quantile sketch family * The KLL quantile sketch family * The REQ quantile sketch -The one empirical quantile sketch is the T-Digest sketch. +The one empirical quantile sketch is the T-Digest sketch. -The mathematical error bounds of the Classic, KLL and REQ sketches are specified with respect to rank and not with respect to quantiles. +The mathematical error bounds of the Classic, KLL and REQ sketches are specified with respect to rank and not with respect to quantiles. The T-Digest is empirical and has no mathematical basis for estimating its error and its results are dependent on the input data. However, for many common data distributions, it can produce excellent results. Please refer to the spcific documentation about the T-Digest sketch. For the Classic and KLL sketches, the difference between the rank upper bound and the rank lower bound is a 99% confidence interval and is an additive constant for all normalized ranks between 0.0 and 1.0. The specific error is a function of the parameter <i>K</i> of the sketch and can be derived from the sketch. For example, if the rank error for a given K is 1%, then the error of a result rank of .01 is +/- .01 with a 99% confidence; the error of a result rank of .99 is +/- .01 with a 99% confidence. -The REQ sketch is special in that it's error is also relative to the actual result rank (thus its name: Relative Error Quantiles). It was designed to proved very high rank accuacy for either the high end of the range of ranks (close to 1.0) or, based on the user's choice, the low end of ranks (close to 0.0). Please refer to the spcific documentation about the REQ sketch. +The REQ sketch is special in that its error is also relative to the actual result rank (thus its name: Relative Error Quantiles). It was designed to proved very high rank accuacy for either the high end of the range of ranks (close to 1.0) or, based on the user's choice, the low end of ranks (close to 0.0). Please refer to the spcific documentation about the REQ sketch. Review Comment: `It was designed to proved` seems incorrect -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
