Subject: Re: [boost] Formal Review of Proposed Boost.Histogram Library Starts TODAY
From: Hans Dembinski (hans.dembinski_at_[hidden])
Date: 2018-09-26 16:23:26
> On 26. Sep 2018, at 15:07, <a.hagen-zanker_at_[hidden]> <a.hagen-zanker_at_[hidden]> wrote:
>>> "value()" seems to be a misnomer, since it is used to print a label for the bin (in your second example: getting_started_listing_02.cpp line 41). To me it is not logical for a bin to have a value() member function, since a bin really is a domain of values.
>
>> That depends on the axis. axis::category has bins which represent a single value, not a continuum.
>
> A single value, a set of values and a continuum of values are all domains of values. A bin logically is a domain of values.
Histogram uses a generalized bin concept, where a single bin can represent a single value on an axis or a domain of values. You cannot implement a category axis with your narrow concept of bin. When you have an category axis that represents the sequence "red", "green", "blue", then what's the domain for the bin "red"?
>> That's the example you are referring to. Same for axis::integer. The other axes types use a different kind of bin type that represents an interval. That's why there are two helper templates interval_view and value_view. Use the former for axis types whose bins represent an interval and the latter for axis types whose bins represent a single item.
>
> I don't agree, a bin always is a domain of values and there are more options than interval and value. For instance, I could chose to have bins for odd and even integers. I think you need an interface for the generic case: a bin is a domain.
You are free to define your bin_type in almost any way you like.
>>> operator int() seems inappropriate because why would a bin need to
>>> know its index? That should logically be the responsibility of the
>>> axis
>
>> I explained the benefit in an answer to Steven.
>
> But, I don't think that convenience is sufficient reason. It is not logical and brings about further questions. A histogram is a relatively simple mathematical construct, it must be possible to support it with concepts that are rigorously logical, not just convenient.
Why is it illogical that a bin knows its own index? A bin both represents a location on an axis and a value or value range on that axis.
The library is not forcing you to define your own bins in this way.