2011/9/15 Reto Bachmann-Gmür <[email protected]>:
> On Thu, Sep 15, 2011 at 3:52 PM, Rupert Westenthaler
> <[email protected]> wrote:
>> On Thu, Sep 15, 2011 at 2:31 PM, Reto Bachmann-Gmür <[email protected]> wrote:
>>> Hello
>>>
>>> In the version of sling-stanbol[1] I've just pushed to github I'm
>>> adding all the enhancement to an MGraph. The idea is to allow queries
>>> on all the enhancements extracted from the jcr content. An issue is
>>> avoiding redundant information in the rdf store and the following
>>> questions arised:
>>>
>>> - Why do the resources of type fise:Enhancement have URIs? Do we
>>> expect that (rest) clients want to reference to such a resource?
>>>
>>
>> In case you want to provide feedback this might be the case.
> The training api I'd like to see would rather be stateless.

It's probably not possible to have a stateless training API for most
statistical models: to train a model (e.g. for OpenNLP Named Entity
detection or text categorization using) you typically need thousands
of manually annotated documents: you cannot send them as one query to
the server: you need to upload them incrementally one after the other,
refine  the annotations and from time to time re-send a query to
update the model from the previously uploaded and annotated documents.

-- 
Olivier
http://twitter.com/ogrisel - http://github.com/ogrisel

Reply via email to