Hi Egon,

On 30 October 2010 11:06, Egon Willighagen <egon.willighagen at gmail.com>wrote:

> Hi Nina,
>
> On Sat, Oct 30, 2010 at 8:34 AM, Nina Jeliazkova
> <jeliazkova.nina at gmail.com> wrote:
> > - Should it be exposed by OpenTox services as ot:Algorithm or ot:Model ?
>
> Is a ot:Model a subclass of ot:Algorithm?
>
>
No,  not in the current version of opentox.owl .



> Then again... ot:Algorithm is an implementation of an algorithm,
> correct?


Well, not really, it's actually a REST resource that is applied to get
something done (create a model,  calculate descriptors), not an abstract
reference to an algorithm.  In opentox.owl there is no notation of
"implementing algorithm" , as in Blue Obelisk dictionary.


And an ot:Model would be a more abstract 'model'... but
> ot:Model is actually the implementation too, right?


The general idea was that a Model is generated by an Algorithm, given some
data and parameters . This comes mostly from statistical/machine learning
algorithms - there is some data, one applies the algorithm of "linear
regression" or "support vector machines" and receives a model that could be
applied for prediction (e.g. linear regression equation , or something more
complex in case of SVM).



> Otherwise, you
> could not use it to actually have it calculate something...
>
>
I am afraid we don't have defined "being able to calculate" property in
opentox.owl , this is defined at the REST level.   Would be good to
have/merge REST operations and the ontology indeed.


> > - What is the right way to use / extend Blue Obelisk descriptors
> dictionary
> > to describe this implementation?
>
> At the Blue Obelisk Descriptor ontology, we do not have the concept of
> a 'model', but algorithms (descriptions or protocols for how to do
> something), and implementations (algorithms expressed in a programming
> language, possibly using a particular set of data, such as isotope
> masses, atom type lists, etc).
>

Do you have idea if it is possible to describe machine learning algorithms
via BO ontology ?  Or we need just to adopt a different ontology for data
mining algorithms.  Models are the ultimate objects coming from OpenTox
framework, they are being built and validated, this seems a bit different
from algorithms in BO.

Nina


>
> Egon
>
> --
> Dr E.L. Willighagen
> Postdoctoral Research Associate
> University of Cambridge
> Homepage: http://egonw.github.com/
> LinkedIn: http://se.linkedin.com/in/egonw
> Blog: http://chem-bla-ics.blogspot.com/
> PubList: http://www.citeulike.org/user/egonw/tag/papers
> _______________________________________________
> Development mailing list
> Development at opentox.org
> http://www.opentox.org/mailman/listinfo/development
>