Design issues for NeXus for errors and treated data

[Mark Koennecke]

High everybody,

This is also partly a repsonse to Tanya's earlier message about
errors.

On Wed, 28 Mar 2001, Tanya Maria Riseman wrote:

> 	Unless the storage of "treated" data and the results from data reduction
>  is done in a SUPERIOR way, there is almost no point in doing at all, because 
>  everyone has their own favorite "good-enough" programs like Origin. 

   I think the SUPERIOR is the keyword here. Of course it is nice to have
   a data analysis tool which logs which files were used, what has been
   done to the data etc. Such a beast does not yet exist. Or do you
   start writing one, Tanya, now that you have time? If somebody starts
   on such a job we surely can work together with the person in order to
   add necessary goups and features to NeXus. However, it might not be a
   good idea to define a standard in advance and thus force design 
   decisions on someone starting on such a data analysis tool. 

   Moreover the number of instruments correlated with specific data
   analysis methods gives rise to an almost infinite number of
   combinations. May be this is to much work for us to define it all.

> 	I think that it is better to keep the NeXus defintion completely 
>  separate from the reduced and treated data defintion ("Ratdata"?).
>  Maybe 3 sections: NeXus, Ratdata and code/tools common to both.

   OK.

   Concerning errors:

   I do not like the idea of folding error information into the data. Two
   reasons: Own data types can again only be defined in HDF-5. We loose
   compatibility with HDF-4. Second: if somebody wishes to plot the data 
   without errors she should not need to unravel sophisticated data
   structures. I am afraid that NeXus will not be used if it is
   to complicated to get down to the data. However it should be
   easy to find the errors if someone wants to do a decent job and plot
   the errors with the data. I opt for a scheme in which the errors are
   stored in the same vGroup as the original data but with _error 
   appended to the name. For example:

	counts[]
        counts_error[]

        xaxis[]
	xaxis_error[]
 	
   This is just my 2c worth of opinion.

	Have a nice weekend

			Mark Koennecke