Re: [boost] Timeseries review

From: Eric Niebler (eric_at_[hidden])
Date: 2007-08-06 13:30:59

Sanjeev Mohindra wrote:
> Just doing through it now. It would be very useful. Some quick questions and
> comments based on very brief perusal.
> 
> 1.  How does one handle missing (time-series) values in this framework? So,
> for example, I have set up my time-series with a certain discretization and
> then at some point(s) along the series, I receive no data, what would be an
> elegant way of dealing with this situation in the library. We would also
> need a policy to ignore or interpolate the values of missing data.

There are a couple of options. One would be to use a sparse time series, 
and only insert (value,offset) elements for the data points you receive. 
Another option would be to pick a "magic" number and use that in the 
place of missing data points. If the data is dense otherwise, this would 
be the most space-efficient way to store the data. Either way, you'll 
need a pass over the data to interpolate the missing data.

> 2. I like the resampling support (Coarse_grain and Fine_grain)

Interpolation is one resampling strategy that I would very much like to 
add. It's a common use, and the library should support it.

> 3. Did I say, the library would be extremely useful.

Thanks!

-- 
Eric Niebler
Boost Consulting
www.boost-consulting.com
The Astoria Seminar ==> http://www.astoriaseminar.com