Re: [Boost-users] [math-toolkit] algorithms for"selfmade"distributions

From: Christian Henning (chhenning_at_[hidden])
Date: 2008-06-26 12:05:54

John, all documents I have are internal company documents. I wasn't
able to find a good website for what I want. But we use Intel's IPP
and it is providing the function ippsConv_64f which does the trick.
The reason I would like to use the Math-Toolkit is that we have
humongous preprocessing before finally calling that function. For
instance:

* one of the two distibution to be convolved is empty or has just one point
* the two distribution need to have the same delta

In the end we just have some pdf's that need to be convolved for
further processing. Given the nature of the beast they can have all
kinds of legal shapes.

In going forward, I think having a data structure for holding
"selfmade" pdf's would be a big plus. I don't mind working on it, as
long as, I have some guidance.

Regards,
Christian

On Wed, Jun 25, 2008 at 4:37 AM, John Maddock <john_at_[hidden]> wrote:
>
>> Yes, more than just the weighted mean. What would need to change
>> inside a pdf object when convolving two distributions? As far as I
>> understand, and please correct me if I'm wrong, the pdf's are
>> lightweight in a sense that every y-value is being computed on the fly
>> based on the x-value?
>
> Yes, the result is computed numerically each time the pdf or cdf etc
> functions are called, however the interface is just:
>
> result_type pdf(distribution_object, random_variable);
>
> So as long as there is some mechanism for calculating the result from the
> convolved distributions then the interface should be able to cope?
>
> Do you have a reference/web page I can look at for what you want to
> calculate so I can see what's involved?
>
> John.
> _______________________________________________
> Boost-users mailing list
> Boost-users_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/boost-users
>