[HN Gopher] Linear-log bucketing: fast, versatile, simple (2015)
___________________________________________________________________
Linear-log bucketing: fast, versatile, simple (2015)
Author : signa11
Score : 35 points
Date : 2023-01-24 09:23 UTC (13 hours ago)
(HTM) web link (pvk.ca)
(TXT) w3m dump (pvk.ca)
| JensRantil wrote:
| Sounds very similar to how memcached allocates slabs.
| sbaiddn wrote:
| Linear-log -> asinh(x)
|
| Its linear for small values and log for large. It also has the
| advantage that it is continuous and smooth. (Obviously asinh is
| not very useful for allocation)
|
| I first thought of it while pondering an amp that would best use
| the limited dynamic range of a (analog) recording media. Thats
| what I do in my free time.
|
| I later realized that asinh(f(x)) is a _great_ substitute for log
| space when plotting. It preserves the sign and the small value
| details while compressing large values. Unfortunately no one os
| familiar with this mapping.
|
| EDIT: from apeheele's comment below I noticed I made a mistake
| and its not tanh, but asinh that is linear for small x and
| logarithmic for large c.
| apwheele wrote:
| One I have seen for time-series forecasting is the inverse-
| hyperbolic sine function as well,
| https://robjhyndman.com/hyndsight/transformations/.
| dang wrote:
| Discussed at the time:
|
| _Linear-Log Bucketing: Fast, Versatile, Simple_ -
| https://news.ycombinator.com/item?id=9792184 - June 2015 (8
| comments)
| fanf2 wrote:
| Paul Khuong encouraged me to use this code to make my hg64
| histogram bucketing faster:
| https://dotat.at/@/2022-10-12-histogram.html
___________________________________________________________________
(page generated 2023-01-24 23:02 UTC)