[SciPy-user] TimeSeries - Questions Re lib/moving_funcs.py

[Pierre GM]

On Wednesday 05 December 2007 00:47:03 pablo mitchell wrote:
> The C source code seems able to accommodate my suggestion of making the
> covariance the central function and the remaining ones the facades.  Just a
> suggestion.

Mmh, you know, I can think about something that would be even better than a 
suggestion ;). Feel free to start implementing something, we could definitely 
use some more hands !

>
> Another suggestion:  a new class similar to TimeSeriesRecords.
>
> For my purposes, I would prefer to be able to instantiate a 2-D time-series
> that can be indexed not just on dates (temporal) but also on variable names
> (spatial if you will).  

Yeah, I'd like to have named columns on an homogeneous 2D array as well. I 
guess that could be doable with an extra dictionary (column position:field 
name), and a tailored __getslice__. But I'd see that more as a numpy 
extension, as an hybrid between a 2D ndarray and a record array. 
However, would it be really efficient ? I have the nagging feeling that the 
current method for slicing on dates can be improved.