Re: [h5md-user] description attribute

[Pierre deBuyl]

On Wed, Aug 17, 2011 at 09:07:33PM +0200, Felix Höfling wrote:
> Am 17.08.2011, 19:08 Uhr, schrieb Peter Colberg  
> <address@hidden>:
> 
> >On Wed, Aug 17, 2011 at 06:23:59PM +0200, Felix Höfling wrote:
> >>Hi Pierre and Peter,
> >>
> >>I suggest an optional attribute "description" to each data set.
> >>Descriptive group identifiers are useful, but sometimes not precise
> >>enough. For example, I would like to have a description like
> >>"potential energy per particle" attached to the observable group
> >>potential_energy.
> >>
> >>Are there any objections to this?
> >
> >I find that short ASCII descriptions tend to add little additional
> >value, e.g. “potential energy per particle” clarifies that the energy
> >is per particle, but does not specify the unit. Further I would rather
> >have a mathematical formula as a description of a dataset, human
> >languages are often too imprecise for this purpose.
> >
> >As such I would rather keep descriptions out of the data file, and
> >include them in the documentation of a simulation package.
> >
> >We already store a h5md/creator_version attribute, and we encourage
> >users to use long, descriptive dataset (and group) names such as
> >“potential_energy” (instead of “EPOT” or “en_pot”), so it is 
> >easy
> >to look up a detailed, exact description with formula(s) in the
> >documentation.
> >
> 
> I agree with the point on reading the documentation. This can not be  
> replaced by a description field, of course.
> 
> But for less common observables, such a field might be useful. A practical  
> example: How should I name the group containing the total energy of the  
> heat bath variables in a Nosé-Hoover chain? I think  
> total_energy_nose_hoover is concise, but might be pretty obscure to  
> somebody not familiar with the details of the creator software.
> 
> Further for output purposes, a description field would easy the  
> translation of a group name into a descriptive string using human language.
> 
> Felix
> 
I would prefer to consider such additions for after the first release of H5MD.
For now, we should concentrate on a working core set of features. Besides, 
adding these attributes to your datafiles would not break compliance with 
the H5MD specification.