[
https://issues.apache.org/jira/browse/MAHOUT-676?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13040750#comment-13040750
]
Ted Dunning commented on MAHOUT-676:
------------------------------------
This is a bit of an odd use of a slice sampler.
Normally slice samplers are used in the sense that Radford Neal proposed in his
2003 (I think) paper. That is,
they are used to sample from a distribution that is contained in R^n and were
you know the PDF times a possibly
unknown constant. The simplest case is the univariate case where you start
with some x in the domain of the
distribution, sample y ~ Uniform[0, p(x)] and then sample uniformly from X = {x
st p(x) < y}. For unimodel p, this
is very nice.
For sampling from a finite set the sampling from X involves some kind of search
which makes slice sampling not much
better than other fast multinomial samples. For instance, you can build a
Huffman or arithmetic encoder for the
different symbols to be sampled. Then you just decode random bits until you
have a unique result. This gives you
log(n) speed for sampling. Similarly (and effectively the same), you can use
bisection until you get a unique result.
What is the need being satisfied here?
> Random samplers in a modular library
> ------------------------------------
>
> Key: MAHOUT-676
> URL: https://issues.apache.org/jira/browse/MAHOUT-676
> Project: Mahout
> Issue Type: New Feature
> Components: Math
> Reporter: Lance Norskog
> Priority: Minor
> Attachments: MAHOUT-676.patch, Sampler.patch
>
>
> This is a modular suite of samplers.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
