Hi Dave, my answers and comments are in the following. It’s good to talk about RETE engines :-)
> Le 28 janv. 2015 à 19:33, Dave Reynolds <[email protected]> a écrit : > > On 28/01/15 14:06, Christophe FAGOT [intactile DESIGN] wrote: >> Hi Andy, >> >> thanks for your answer, and I’m ok for the graph being a set of triples, it >> is the (very good) reason explaining why only one triple is produced. But >> the reasoner is not in forward model. It is a forward-RETE model, which >> means that the forward rules have to work incrementally, allowing to add and >> remove triples and maintaining the consistency of the model. >> >> So in the case described by Sébastien, the forward-RETE model should not >> remove the inferred triple since another rule has its body terms still >> validated. At least, this last rule should have been fired in order to >> indicate it that the triple which was not created previously (because it was >> still in the graph) is going to be removed, so this last rule should produce >> it again. > > The RETE engine stops once a triple has been deduced by one route. If you > attempt to track each possible route by which a triple could be deduced and > reference count them all then you will get a combinatoric explosion in > numbers of possible deduction paths and performance plummets (which is why > naive truth maintenance never worked out). You don’t need to track each possible route to maintain the truth. In a previous company Sébastien and I created a home-made RETE engine, and simple counters stored with triples are enough to maintain the truth of the graph. The counter is increased each time a triple is redundantly added, and decreased when the triple is asked to be removed. The triple is really removed from the graph when its counter is at 0. So, no combinatoric explosion. > > The Jena engine works around this by not attempting to handle removals > incrementally at all. A remove is supposed to mark the model as needing a new > "prepare" stage and the entire deduction process is run from scratch again > the next time you query the model. So, if I well understand what you say, the RETE Engine is monotonic and maintain the truth when adding some triples, but it’s not the case when triples are removed. Am I true? If, when removing a triple, we want to maintain the truth in a graph managed by the RETE engine, we have to do as if the engine was a classical forward one and require a new « prepare » stage (meaning the removing of the whole previous inferences). Am I true on this point too ? If what I write is true, I now fully understand what you wrote to me in the past about the removal into the RETE engine :-D > > That certainly used to work and I can't see why Sébastien's case would fail, > though I don't see the code by which the results are getting accessed. I'm > not in a position to test it from here. It fails if, for whatever reason depending on the software you are developing, you can’t remove the whole infered triples each time you remove a single one.And it’s our case :-( But whatever, having your explanations is good for us to understand if and where we can patch the current Jena RETE engine (and dependencies), or if we have to deal with it by our side. Chris. > > Dave > >> Chris. >> >> Christophe FAGOT, PhD >> RESPONSABLE R&D INFORMATIQUE >> >> intactile DESIGN >> Création d’interfaces + subtiles >> +33 (0)4 67 52 88 61 >> +33 (0)9 50 12 05 66 >> 20 rue du carré du roi >> 34000 MONTPELLIER >> France >> www.intactile.com <http://intactile.com/> >> >> Hugh MacLeod : "It's not what the software does, it's what the user does" >> >> Les informations contenues dans cet email et ses documents attachés sont >> confidentielles. Elles sont exclusivement adressées aux destinataires >> explicitement désignés ci-dessus et ne peuvent être divulguées sans >> consentement de son auteur. Si vous n'êtes pas le destinataire de cet email >> vous ne devez pas employer, révéler, distribuer, copier, imprimer ou >> transmettre le contenu de cet email et devez le détruire immédiatement. >> >>> Le 28 janv. 2015 à 12:17, Andy Seaborne <[email protected]> a écrit : >>> >>> (Dave is not around at the moment so I'll try to answer some parts of your >>> question ...) >>> >>> On 28/01/15 10:28, "Sébastien Boulet [intactile DESIGN]" wrote: >>>> Hello, >>>> >>>> I have two rules which could produce the same triple: >>>> >>>> String rules = "[r1: (?a eg:p ?b) -> (?a, eg:q, ?b)]" + >>>> "[r2: (?a eg:r ?b) -> (?a, eg:q, ?b)]"; >>>> >>>> i have configured a GenericRuleReasoner in FORWARD_RETE mode. >>>> >>>> GenericRuleReasoner reasoner = new >>>> GenericRuleReasoner(Rule.parseRules(rules)); >>>> reasoner.setMode(GenericRuleReasoner.FORWARD_RETE); >>>> InfModel model = ModelFactory.createInfModel(reasoner, >>>> ModelFactory.createDefaultModel()); >>>> >>>> When a triple satisfy the first rule and another triple satisfy the second >>>> rule: >>>> >>>> Resource subject = model.createResource(); >>>> Property predicateP = model.getProperty("urn:x-hp:eg/p"); >>>> Literal literalA = model.createTypedLiteral("A"); >>>> Property predicateR = model.getProperty("urn:x-hp:eg/r"); >>>> >>>> model.add(subject, predicateP, literalA); >>>> model.add(subject, predicateR, literalA); >>>> >>>> only one triple is deduced: >>> >>> An RDF graph is a set of triples. >>> >>> A set only has one of each thing in it. >>> >>> If you >>> add(triple) >>> add(triple) >>> >>> you will see only one triple in the output. This is not to do with >>> inference, it is to do with an RDF graph being a set. >>> >>>> >>>> <rdf:RDF >>>> xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"; >>>> xmlns:j.0="urn:x-hp:eg/" > >>>> <rdf:Description rdf:nodeID="A0"> >>>> <j.0:r rdf:datatype="http://www.w3.org/2001/XMLSchema#string";>A</j.0:r> >>>> <j.0:p rdf:datatype="http://www.w3.org/2001/XMLSchema#string";>A</j.0:p> >>>> <j.0:q rdf:datatype="http://www.w3.org/2001/XMLSchema#string";>A</j.0:q> >>>> </rdf:Description> >>>> </rdf:RDF> >>>> >>>> When i remove the fist triple: >>>> >>>> model.remove(subject, predicateP, literalA); >>>> >>>> the sole deduced triple is removed even if the second rule is still >>>> satisfied: >>> >>> You ran the reasoner in forward model - it included all deductions at the >>> start and then does not run again until you ask it to. >>> >>> To trigger it again: >>> >>> InfModel.rebind() >>> "Cause the inference model to reconsult the underlying data to take into >>> account changes." >>> >>> or run in backward mode. >>> >>> Andy >>> >>>> >>>> <rdf:RDF >>>> xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"; >>>> xmlns:j.0="urn:x-hp:eg/" > >>>> <rdf:Description rdf:nodeID="A0"> >>>> <j.0:r rdf:datatype="http://www.w3.org/2001/XMLSchema#string";>A</j.0:r> >>>> </rdf:Description> >>>> </rdf:RDF> >>>> >>>> is it the expected behavior ? >>>> is there a workaround to deduce twice the same triple or at least to don’t >>>> remove the sole deduction ? >>>> >>>> Thanks >>>> >>>> >>>> Sébastien BOULET >>>> LEAD DÉVELOPPEUR >>>> >>>> intactile DESIGN >>>> Création d’interfaces + subtiles >>>> 04 67 52 88 61 >>>> 09 50 12 05 66 >>>> 20 rue du carré du roi >>>> 34000 MONTPELLIER >>>> France >>>> www.intactile.com <http://intactile.com/> >>>> >>>> Les informations contenues dans cet email et ses documents attachés sont >>>> confidentielles. Elles sont exclusivement adressées aux destinataires >>>> explicitement désignés ci-dessus et ne peuvent être divulguées sans >>>> consentement de son auteur. Si vous n'êtes pas le destinataire de cet >>>> email vous ne devez pas employer, révéler, distribuer, copier, imprimer ou >>>> transmettre le contenu de cet email et devez le détruire immédiatement. >> >> >
