Re: [boost] [ANN] Boost.Time_series: feedback requested

From: Eric Niebler (eric_at_[hidden])
Date: 2006-12-14 20:01:13

Austin Bingham wrote:
> Like many other posters, I'm excited to see this work being
> done. Not to be shown up by Zach, I also have a few questions
> about the library.
> 
> First, it's not entirely clear to me whether arithmetic
> operations between series with different non-integer discretizations
> is allowed. From the documentation:
> 
>   Arithmetic operations between series is only defined when the 
>   series have the same discretization. When the discretization 
>   is a plain integer, the discretizations are checked for 
>   compatibility at runtime.
> 
> If I'm reading this correctly, arithmetic between unequal 
> non-integer series is undefined but not checked at runtime. If
> so, why isn't it checked?

It is checked. That's a doc bug. It should read, "When the 
discretization is not a compile-time constant, the discretizations are 
checked for compatibility at runtime."

> Along those lines, I can envision a situation where I *would*
> want to do the kind of arithmetic described above. In some
> applications, the notions of time and range are effectively
> interchangeable. If I have two "time series", one discretized
> on range and one on time but both representing the same
> physical space, I might reasonably expect to be able to 
> add them together. I realize I'm stretching a bit here,
> but I want to get your sense of how things like this might
> be dealt with in this library.

I'm afraid I don't follow. What do you mean by a time series 
"discretized on range"?

> Does the clip() function necessarily copy the source data? It's
> often nice to be able to generate a sliced view of some source
> data without actually copying it. I assume clip()
> does copy, so I was wondering if there is a facility for
> prodcing non-copying views of a time series.

clip() does *not* copy the data. It returns a clipped view. If you want 
to copy the data, you should assign the clipped view to another series.

> Again, I really like the work you've done so far, and I'd love
> to have a robust, high-feature time-series library. I'll keep
> looking things over and let you know if I've got any more
> questions.

Thanks. Keep the questions coming.

-- 
Eric Niebler
Boost Consulting
www.boost-consulting.com