On 10/11/15 20:13, Juan Sequeda wrote:
Andy,
On Tue, Nov 10, 2015 at 3:06 AM, Andy Seaborne <[email protected]> wrote:
On 10/11/15 03:25, Juan Sequeda wrote:
Rob, all
The query does work for my needs. Thanks!
Each Tree graph will consist of 10-20 triples for a given root. Given
those
circumstances, what do you think performance would be?
As I mentioned, another idea is to put the RDF graph in a NG and then just
clear that NG.
What is the experience of using a high number of Named Graphs with small
number of triples in each named graph (~10-20 triples). The
operations/queries I am expecting to do are:
- Given a root, return me the tree graph associated to that root
CONSTRUCT {?s ?p ?o} { GRAPH <http://example/> { ?s ?p ?o } }
- Given a root, delete the tree graph associated to that root.
CLEAR GRAPH <http://example/>
- Insert a graph
INSERT DATA { GRAPH <http://example/> { ... } }
Thanks for the pointers.
Hi Juan,
Exact details depend on the backend. I guess you wil then use default
union graph to query all the NG's as one?
In my case, I do not need to query all the NGs at once. My query workload
will always have the NG uri.
For the current general purpose in-memory dataset, GRAPH ?g is a loop as
would be a default union graph, so huge numbers may be an issue for query
over all graphs, but of course it's in-memory so limited in scale anyway.
For TDB, a graph name is just a column and default union graph should be
as fast.
I am using TDB. So looks like I'm ok.
For new fully transactional dataset, not yet released (JENA-624), the
design allows for default union graph in a manner similar to TDB.
Cool to know.
Btw, in your opinion, what would be the best way to go? Use the SPARQL
property path query that Rob suggested or use the NG approach? I'd truly
appreciate your insight.
Personal opinion - the Graph Store Protocol is a great way to manage the
dataset. That's the NG way + using the HTTP verbs to PUT (replace),
POST (append) and DELETE graphs.
There is an equivalence between GSP and certain SPARQL Update management
operations so choose the way that fits your workflow.
The one point to recognize is that a dataset can only use named graph
for one thing. And we onyl have quads ... not quintuples or more.
Andy
Thanks!
Andy
--
Juan Sequeda, Ph.D
+1-575-SEQ-UEDA
www.juansequeda.com
On Mon, Nov 9, 2015 at 6:12 PM, Rob Vesse <[email protected]> wrote:
Juan
I haven't tested this but in principal the following should work:
DELETE
{
?s ?p ?o
}
WHERE
{
{
# Find things directly connected to the root
<root> ?p ?o .
BIND(<root> AS ?s)
}
UNION
{
# Find everything indirectly connected to the root
<root> (<>|!<>)+ ?s .
?s ?p ?o .
}
}
In the first branch simply select any triples directly off of the root.
In the second branch find any triples one/more steps away from the root
and the associated triples. Then in the DELETE template simply delete
all
the matched triples.
The second branch uses Joshua Taylor's trick from
http://stackoverflow.com/a/26707541/107591 of using a property path via
a
URI and its negation to find any subject that is reachable by 1 or more
steps via any property from your root. You can then simply grab things
that are directly connected to that subject.
This will probably perform badly and can be sped up if you have a
specific
property used to denote the tree hierarchy. For example if the root is
linked to child nodes via the property :child then we could rewrite the
second branch as follows:
<root> :child+ ?s .
?s ?p ?o .
Which still requires a property path but will perform better because the
path is simpler.
Rob
On 09/11/2015 15:32, "Charles Greer" <[email protected]>
wrote:
Hi Juan,
Seems that if you can do this efficiently
"Another thought is to put the RDF graph in a NG and then just clear the
NG."
then that will be more efficient than following property paths.
Interesting question to be sure.
Charles
________________________________________
From: Juan Sequeda [[email protected]]
Sent: Monday, November 09, 2015 3:28 PM
To: [email protected]
Subject: Deleting an entire RDF graph
All,
Assume I have an RDF graph which is a tree, hence there is a root node.
What is the best way of deleting that entire RDF graph given the root
node?
Is there a way to do it with Property paths? And if so, I guess I would
have to have knowledge of the structure of the graph/tree?
Another thought is to put the RDF graph in a NG and then just clear the
NG.
Thoughts?
Thanks,
Juan
