You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@commons.apache.org by "Alex Herbert (Jira)" <ji...@apache.org> on 2021/07/08 08:51:00 UTC
[jira] [Commented] (RNG-150) Update samplers to use the ziggurat
exponential sampler
[ https://issues.apache.org/jira/browse/RNG-150?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17377221#comment-17377221 ]
Alex Herbert commented on RNG-150:
----------------------------------
This change also applies to the GeometricSampler. Here is the current performance when using the AhrensDieterExponentialSampler:
{noformat}
Benchmark (probabilityOfSuccess) (randomSourceName) (samplerType) Mode Cnt Score Error Units
GeometricSamplersPerformance.baseline N/A N/A N/A avgt 5 2.274 ± 0.010 ns/op
GeometricSamplersPerformance.sample 0.1 XO_RO_SHI_RO_128_PP GeometricSampler avgt 5 28.249 ± 0.101 ns/op
GeometricSamplersPerformance.sample 0.1 XO_RO_SHI_RO_128_PP InverseTransformDiscreteSampler avgt 5 52.946 ± 0.129 ns/op
GeometricSamplersPerformance.sample 0.3 XO_RO_SHI_RO_128_PP GeometricSampler avgt 5 29.253 ± 0.442 ns/op
GeometricSamplersPerformance.sample 0.3 XO_RO_SHI_RO_128_PP InverseTransformDiscreteSampler avgt 5 51.363 ± 0.115 ns/op
{noformat}
Here is the performance after changing to ZigguratSampler.Exponential:
{noformat}
Benchmark (probabilityOfSuccess) (randomSourceName) (samplerType) Mode Cnt Score Error Units
GeometricSamplersPerformance.baseline N/A N/A N/A avgt 5 2.982 ± 0.018 ns/op
GeometricSamplersPerformance.sample 0.1 XO_RO_SHI_RO_128_PP GeometricSampler avgt 5 11.074 ± 0.013 ns/op
GeometricSamplersPerformance.sample 0.1 XO_RO_SHI_RO_128_PP InverseTransformDiscreteSampler avgt 5 52.994 ± 0.110 ns/op
GeometricSamplersPerformance.sample 0.3 XO_RO_SHI_RO_128_PP GeometricSampler avgt 5 13.118 ± 0.041 ns/op
GeometricSamplersPerformance.sample 0.3 XO_RO_SHI_RO_128_PP InverseTransformDiscreteSampler avgt 5 51.416 ± 0.391 ns/op
{noformat}
Approximately 2.5x faster.
> Update samplers to use the ziggurat exponential sampler
> -------------------------------------------------------
>
> Key: RNG-150
> URL: https://issues.apache.org/jira/browse/RNG-150
> Project: Commons RNG
> Issue Type: Improvement
> Components: sampling
> Affects Versions: 1.3
> Reporter: Alex Herbert
> Assignee: Alex Herbert
> Priority: Trivial
> Fix For: 1.4
>
>
> The LargeMeanPoissionSampler uses a standard exponential deviate.
> The GeometricSampler uses an exponential deviate.
> These can be changed from the AhrensDieterExponentialSampler to the ziggurat exponential sampler which is 4x faster.
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)