[
https://issues.apache.org/jira/browse/MATH-1293?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15011124#comment-15011124
]
Gilles commented on MATH-1293:
------------------------------
It's a step in the right direction but at this point, it's only my opinion.
You really should raise the issue on the "dev" ML to potentially get other
opinions on how the optimization must be integrated in future versions of CM.
For example, my idea was that the helper class do not need to bother with a
table fixed at compile time; the constructor of {{FastFactorialLog}} would
indeed initialize everything according to the caller's request (e.g. cache
size).
It would be the role of {{CombinatoricsUtils}} to decide how to instantiate a
{{FastFactorialLog}} object; and the policy (e.g. lazy instantiation at the
first call to {{factorialLog}} or in a _static_ block) must be discussed on the
"dev" ML.
Since we want to have more flexible interfaces, we might want to test
alternative design for those "small" utilities.
An idea would be to create a function object:
{code}
FastFactorialLog fact =
CombinatoricsUtils.createFactorialLog().withCacheSize(1024);
{code}
Then the caller could use the returned object:
{code}
double r = fact.value(123);
{code}
So a user won't depend on how we decided to implement the {{factorialLog}}
static function (and maybe the function should be deprecated if the same
functionality and more can be achieved in a better way).
There might be other suggestion of usability improvements or changes... So,
before you embark on yet another patch, you should really raise the issue on
the ML. Thanks!
> Tabulating the logarithmic factorial
> ------------------------------------
>
> Key: MATH-1293
> URL: https://issues.apache.org/jira/browse/MATH-1293
> Project: Commons Math
> Issue Type: Improvement
> Affects Versions: 3.5
> Reporter: Aleksei Dievskii
> Priority: Minor
> Labels: feature, patch
> Attachments: factlog.patch
>
>
> Presently, CombinatoricsUtils#factorialLog is calculated via a tabulated
> value for the first 21 entries, and for the rest it employs linear-complexity
> direct summing. For the factorial-heavy computations this can be rather
> painful. I propose a patch which allocates additional 16KB of tabulated
> log-factorial values, allowing very fast (constant complexity) computation
> for arguments up to 17000, and delegating to the log-gamma for the rest.
> Benchmarks show a speed-up of up to 200x.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
