Andy,
I had previously been storing the inferenced form of an ontology WITHOUT
transactions. But when I added the use of transactions, I have run into both
out-of-memory issues and significant performance problems.
Is the code below the best way to create a separate inferenced model with use
of transactions?
Dataset dataset = // assume this is initialized properly
try {
dataset.begin(ReadWrite.WRITE);
Model model = dataset.getNamedModel(modelName); // already
stored non-inferenced data
OntModel omodel = ModelFactory.createOntologyModel(spec, model);
omodel.prepare();
Model pomodel = dataset.getNamedModel(inferredName); // this is new
model to contain inferenced data
pomodel.add(omodel); // add inferenced ontology as
data
dataset.commit();
} finally {
dataset.end();
}
Another question:
You refer below to the "backwards rule engine". Could you be specific as to the
OntModelSpec you are referring to?
Or is it simply that OntModel.prepare() is an explicit call to execute the
forward rules and no use of prepare just does backward rules on demand, but
only for what is necessary to answer the particular query?
-----Original Message-----
From: Andy Seaborne [mailto:a...@apache.org]
Sent: Thursday, May 02, 2013 8:50 AM
To: users@jena.apache.org
Subject: Re: transaction and caching
How to best handle inference will depend on how it's being used. There is a
tension between the full rules engine and having database data.
If you can maintain the data in the database only, by infererring once and
writing inferred triples to the database but update is now not automatic.
If that does not work well for you, then you might consider doing inference at
query time with the backwards rule engine.
Normally, the rules engine is hybrid - some forward rules, executed once on
.prepare() then backwards rules on-demand at query time.
When a lot of overhead from the .prepare is going to waste, not used in the
transaction, backwards makes sense, delaying work until needed during
transaction execution.
(Maybe Dave can add something here).
Andy
