Re: [boost] seeking endorsement for histogram library

Subject: Re: [boost] seeking endorsement for histogram library
From: Hans Dembinski (hans.dembinski_at_[hidden])
Date: 2017-11-14 13:24:04

Dear Fabian,

> On 13. Nov 2017, at 18:59, Fabian Bösch via Boost <boost_at_[hidden]> wrote:
> 
> The histogram classes look quite useful, especially the property that
> you can add histograms and thus accumulate data in parallel is
> important in many scientific contexts.

yes, right! :) I also mention that use case in the user guide. :)

> Did you also intend to provide estimates of mean, variance and possibly
> higher order statistical moments of the whole distribution? This can be
> achieved with so-called on-line algorithms (even in the multi-variate
> case) and would help to get more information about your data. This can
> also be implemented such that the summation of histograms still work
> (and probably also the scaling).

assuming I understand you correctly, then I think the functionality of computing moments on-line is already in Boost.Accumulators, see
http://www.boost.org/doc/libs/1_65_1/doc/html/accumulators/user_s_guide.html#accumulators.user_s_guide.the_statistical_accumulators_library

The histogram library has a weak overlap in scope with Boost.Accumulators, but mostly they are complementary. There is a section in the rationale about on this point:

https://htmlpreview.github.io/?https://github.com/HDembinski/histogram/html/doc/html/index.html#histogram.rationale.comparison_to_boost_accumulators

Best regards,
Hans