Dear all,
    Soon our services will be up and running so I'd like to share with
you some extensions as well as some ideas.

1. Rescue Service
    When one applies a DELETE on a resource (such as a model, a task or
a bibtex resource) it is not really deleted from the database but it's
deactivated. The deleted resource will return a status code 404 (not
found) accompanied by an Error Report (in RDF). The deactivated
resources however are still on the server and will remain there for one
month (we have a lot of disk space, so we think we can cope with it...).
Now, if the user wants to recover a resource that was deleter by mistake
he can apply a get on /rescue?id={resource-id} or /rescue?uri={uri of
previously deleted resource}. I was thinking that this can be a
interface with access restricted to system admins only so it remains
easy for the end-user to send an email to us and we can recover the
deleted resource unless one month has passed and it has been deleted
automatically. This can of remain out of specification.

2. BibTeX service
    The BibTeX service the runs under jaqpot is tested however it will
remain inactive or even useless if we don't find a way to associate
BibTeX entries with other resources. I was thinking of a REST API that
would look like the following...

resource  /bibtex/{id}
method   POST
POSTed Parameters 
  1. resource_uri  [ The uri of the resource we want to associate with
the bibtex entry. ]
Header Parameters
  1. subjectid [The authentication token. Specified by the underlying
policy]

This way we establish 1-to-many relationships between bibtex entries and
other resources. Each BibTeX entry will provide a list of the resources
that cite it. Inversely every algorithm and every model will potentially
point to bibtex entry. Algorithms for example will carry a pointer to a
bibliographic reference which the user will be able to download in
various supported formats (such as the standard JSON-based BibTeX
format). From the ontological point of view we need to introduce a new
property, let's say ot:bibtex with domain ot:OpenToxResource and range
knouf:Entry. 


3. Fast RBF Neural Networks
    An initial implementation of a high-performance RBF-NN training
algorithm will be available. It has been tested on a variety of
datasets. 

4. Consensus Modeling

    We currently work on the design of a simple consensus modeling
algorithm. The algorithm accepts as input a series of model URIs and one
of DoA models of the same length (For example, the user has to provide 2
predictive models and 2 corresponding DoA "models" that were trained
using the same dataset. The consensus model that is trained outputs the
average of the predictions that are obtained by the sub-models that
include the submitted instances in their DoA. A cURL request is expected
to look like that:


curl -X POST 
-d model_uri[]={model1}    -d model_uri[]={model2} 
-d doa[]={doa1 for model1} -d doa[]={doa2 for model2} 
host:port/algorithm/consensus 
-H subjectid:{token}

As soon as the services are up again we'll ask from Vedrin to start
smokeping tests and we expect better performance and stability.

Best Regards,
Pantelis