[MOBY-dev] Moby Collections (was: Problems with Biomoby servicesin Taverna 1.2)

[Dirk Haase]

On Wednesday 20 July 2005 11:24, Martin Senger wrote:
> Rebecca,
>    Many thanks for the explanation.
>
>    I think this is the critical part:
> > Basically I think it doesn't make sense to register an output as being a
> > collection if the results don't form an entity!
>
>    I assume that by entity you mean 'set of pieces that belong
> semantically together'.

Exactly.

>    I think that the current concept of Simples and Collections gives you
> full freedom to model your output in any way you consider proper. 

This is actually _not_ the case and that is why we brought up the issue. There 
is no proper way to output things which really make up a 'set of pieces that 
belong semantically together'. For example there is no way to output the 
results of a clustering service (a cluster is a set of sequences, the service 
result is a set of clusters), because a collection of collections is not 
allowed afaik.

Of course one can create a 'cluster object' but this creates rather artificial 
hurdles in workflow creation. Suppose that after the clustering, each cluster 
should be fed into a multiple alignment service. This would presumably take a 
collection of sequences as input. So one would have to decompose the cluster 
object first before it can be further processed.

I think we all agree that on the input side a collection only makes sense if 
it is meant as one semantical entity (example from the API docs: several 
sequences make up the BLAST database that is to be queried). But not so on 
the output side. So we have an inherent unequality which complicates workflow 
building.

> For 
> example, if you do not feel comfortable to put all AGI Codes in one
> collection because they represent semantically different things you can
> split them by types (or whatever semantics you choose) into several
> collections, 

The claim is they are not related...  The only relation is that they pop up 
for the same input but that is obvious for this kind of services and does not 
need to be emphasized by stuffing them into a collection.

>    I believe that we need to keep the concept of Collections open as are
> the general data types in other languages (e.g. hashtables). 

That is a good point, but I'm not sure if it is appropriate to compare the 
concept of Simples/Collections to general data types in programming 
languages.


> I am sorry to 
> bother you with these obvious comments.

No, they are very welcome indeed!

Servus,
dirk