You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@datasketches.apache.org by al...@apache.org on 2023/02/25 01:15:33 UTC
[datasketches-website] branch hll_error created (now 751997ec)
This is an automated email from the ASF dual-hosted git repository.
alsay pushed a change to branch hll_error
in repository https://gitbox.apache.org/repos/asf/datasketches-website.git
at 751997ec clarify
This branch includes the following new commits:
new 751997ec clarify
The 1 revisions listed above as "new" are entirely new to this
repository and will be described in separate emails. The revisions
listed as "add" were already present in the repository and have only
been added to this reference.
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@datasketches.apache.org
For additional commands, e-mail: commits-help@datasketches.apache.org
[datasketches-website] 01/01: clarify
Posted by al...@apache.org.
This is an automated email from the ASF dual-hosted git repository.
alsay pushed a commit to branch hll_error
in repository https://gitbox.apache.org/repos/asf/datasketches-website.git
commit 751997ec6e3b61b04081828f9c954ed9c626e5a1
Author: AlexanderSaydakov <Al...@users.noreply.github.com>
AuthorDate: Fri Feb 24 17:15:13 2023 -0800
clarify
---
docs/HLL/HLL.md | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/docs/HLL/HLL.md b/docs/HLL/HLL.md
index fe536077..97e965a7 100644
--- a/docs/HLL/HLL.md
+++ b/docs/HLL/HLL.md
@@ -86,12 +86,12 @@ The following plot was generated with <i>LgK</i> = 14 using 2<sup>20</sup> trial
The <i>Factor = 0.8326</i> is directly relatable to the Flajolet alpha factor of 1.04.
As a result, this plot demonstrates that this implementation of the HLL sketch
-will be about 20% = (0.8326/1.04 -1) more accurate than a conventional HLL sketch using Flajolet's estimators (or derived estimators).
+will be about 20% = (0.8326/1.04 - 1) more accurate than a conventional HLL sketch using Flajolet's estimators (or derived estimators).
This is partially due to the use of the HIP estimator[1] for range above the transition point, which occurs at about 1500 on the graph.
Below this transition point the accuracy is near zero (an RSE of about 50 ppm), which is far better than any known implementation of HLL.
This is due to a newly developed theory and estimator developed by Kevin Lang[2].
-The base Relative Standard Error (RSE) for this sketch (at LgK = 14) is 0.0065 = 0.8326 / 2<sup>7</sup>.
+The base Relative Standard Error (RSE) for this sketch (at LgK = 14) is 0.0065 = 0.8326 / sqrt(2<sup>14</sup>).
The horizontal gridlines are configured to be +/- multiples of the base RSE.
The different color curves are contours of the actual error distribution measured at normalized rank values
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@datasketches.apache.org
For additional commands, e-mail: commits-help@datasketches.apache.org