Statistical Services

[amedranogil]

new component that offers statistical services over the CHe information.
The services should transform universAAL Restrictions into SPARQL.
So an application can ask for statistical information for a set of individuals, defined as a type expression; plus a property path (to get to the data that needs to be analysed).

will need an ontology for the statistical domain + services.

[cstockloew]

good idea.

As we plan to have SPARQL support in buses, I wonder how much of it may be provided by DataRep and how much should be in the new component. But this should not stop you from starting; we can decide that later.

Do you have some more conrete info? What's the use case? What about the ont? Do we need a new one or can we extend ont.che?

[amedranogil]

The main idea is for applications that try to determine a behaviour out of the context history can reuse these capabilities, after all the behaviour can be over users, or over concrete devices.
So for example the HVAC reasoner can calculate the mean temperature over a period of time (with and without the HVAC activated).

Another interesting feature to add to this (which right now IDK if it is feasible within the SPARQL sintax) is to find the mean time between events. Maybe that is another service all together: that when configured (through a service call with a ContextEventPattern), it automatically searches for and publishes the changes with the last similar event (including the time elapsed). Of course insead of using context events it could through a service.

Back to statistics, there needs to be a new ontology that covers statistical concepts (such as datasets, probabilistic variables, etc..). I also think this is inline with an earlier proposal of having an AI ontology and services as most AI will be based on statistical concepts. Although A quick search in wikipedia, has given me even more ideas opening the doors to other services providing more complex statistical analisys sych as multivaraint analisys (covariance, PCA, etc..) or statistical inference (like regression). That stuff can be left open for thirdparty implementation on the mean time my proposal is more towards the simpler statistical Univariate description .