Re: [boost] Looking for Time Series reviews

From: Matias Capeletto (matias.capeletto_at_[hidden])
Date: 2007-08-08 08:54:54

On 8/7/07, Eric Niebler <eric_at_[hidden]> wrote:
>
> Matias Capeletto wrote:
> > On 8/6/07, John Phillips <phillips_at_[hidden]> wrote:
> >>    To date there have been no reviews of the time series library posted
> >> to the developer or user lists. Is this a bad time for a review, or is
> >> no one interested enough in this library to put in the time to submit one?
> >>    If you feel the library isn't interesting or useful enough, please
> >> post a note to the list so I can see that opinion. If you are struggling
> >> for time to do the review, please post at least a short opinion, and if
> >> a few days extension will make a difference we can investigate that
> >> possibility. However, the next review begins on August 13, so time for
> >> this is somewhat limited.
> >
> > I really want to participate but I gave my last exam last week and now
> > I am heading to an Open Source congress for a week. I think I will not
> > have time for a formal review there :(
> >
> > I have already give Eric feedback a month ago, and I think the library
> > is a nice inclusion for piece wise manipulation series like the ones
> > found in banking account management.
> >
> > I am not convinced that is the best approach for a general discrete
> > signals framework where to build up digital systems libraries, like
> > FIR and IIR filters, filter networks or non linear and adaptive
> > filters.
>
> <snip>
>
> Matias, if you summarize your DSP design ideas for the list, then I can
> tell everyone what I told you ... that a similar design was tried for
> Time_series and found not to work in practice.

I think that for DSP the best way to model a signal is a dense
sequence, that is not based in the RangeRun concepts. I think how to
implement convolution or FIR block processing based on DFT and I do
not find an easy way, but I have not spend enough time on it.

> I asked you to try writing generic algorithms using the concepts you
> designed. Have you?

No, I have not find the time yet. That is why I just said I am not
convinced, no more than that. And I state again that a RangeRun based
library is a good addition to Boost and can cohabit in Boost with
others DSP libraries.

Best regards
Matias

PS: Sorry for the lenght of the messages, I am in the middle of the congress :)