I think to impl this what is needed is a "logical" node. At the moment the entire LHS forms the justification. But if we supported a logical node, we could do this: rule "reachDirect" salience 10 when logical( e : Edge(s1 : source, t1 : target) ) not( Reach(source == s1, target == t1) ) then insertLogical( new Reach(e.getSource(),e.getTarget()) ); System.out.println( "Reach " + e.getSource() + "," + e.getTarget() ); end That means that it would be inserted when there was no Reach, but it would only be retracted when there was no matching Edge. The justification is only for the part of the rule that is in the logical grouping. To do this is actually quite a trivial change in drools, but it's not something we do now. I think one reason why I held off was that i was looking at Jess and Clips that have this and they state you can have multiple logical elements. But i could't figure out how having 2 or 3 would differ, compare to having just one. Anway to support a singe logical element, you'd need to update the parser to support 'logical' conditional element, in the same format as 'not' and 'exists'. Then if you look at RuleTerminalNode you'll see the part of the code that is related to removing the justifications, on a retract or modify - removeLogicalDependencies. Likewise if you look in the DefaultKnowlegeHelper you'll see how the insertion works. That could would instead be copied to the logical node. If a logical node exists the RTN should have an if statement so the same code does not execute again. Any takers? Mark On 19/02/2011 05:20, Simon Chen wrote: > Hi all, > > I know this is kinda an old topic, but I just couldn't get it working. > > Here is a previous attempt using insertLogical() to handle transitive > closure: > http://drools-java-rules-engine.46999.n3.nabble.com/transitive-closure-td... > The problem with this one is that the newly "logically inserted" object > would violate its own "not exists" condition term, thus removing itself, > then goes the infinite circle of insert/remove... > > Here is a post that deals with transitive closure using "insert", but it > doesn't handle object removal correctly: > http://drools-java-rules-engine.46999.n3.nabble.com/one-question-about-Tr... > > > To me, using insertLogical is attractive because it doesn't require me to > write specific rules to handle object removal. Is there a trick that I can > use to actually implement transitive closure with insertLogical? > > Thanks a lot! > -Simon > _______________________________________________ > rules-users mailing list > rules-users(a)lists.jboss.org > https://lists.jboss.org/mailman/listinfo/rules-users > > _______________________________________________ rules-users mailing list rules-users(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/rules-users

Let me make sure I get it correctly... The example you gave seems to be the one-hop case. For the two-hop case, we need something like this when edge(a, b), reach(b, c), not exists reach(a, c) then insertLogical( reach(a,c) ) So, where do you put your logical around? It should include both edge(a,b) and reach(b,c), right?

Another thought, can we have something like testExistsAndInsertLogical() to replace insertLogical()? But this may be buggy, as the conditions are all met, so the rule actually fired... I really need this working, so I'll probably take a crack at modifying drools, although haven't done it before...

Thanks. -Simon On Saturday, February 19, 2011, Mark Proctor<mproctor(a)codehaus.org> wrote: > I think to impl this what is needed is a "logical" node. At the moment > the entire LHS forms the justification. But if we supported a logical > node, we could do this: > rule "reachDirect" > salience 10 > when > logical( e : Edge(s1 : source, t1 : target) ) > not( Reach(source == s1, target == t1) ) > then > insertLogical( new Reach(e.getSource(),e.getTarget()) ); > System.out.println( "Reach " + e.getSource() + "," + > e.getTarget() ); > end > > > That means that it would be inserted when there was no Reach, but it > would only be retracted when there was no matching Edge. The > justification is only for the part of the rule that is in the logical > grouping. > > To do this is actually quite a trivial change in drools, but it's not > something we do now. I think one reason why I held off was that i was > looking at Jess and Clips that have this and they state you can have > multiple logical elements. But i could't figure out how having 2 or 3 > would differ, compare to having just one. > > Anway to support a singe logical element, you'd need to update the > parser to support 'logical' conditional element, in the same format as > 'not' and 'exists'. Then if you look at RuleTerminalNode you'll see the > part of the code that is related to removing the justifications, on a > retract or modify - removeLogicalDependencies. Likewise if you look in > the DefaultKnowlegeHelper you'll see how the insertion works. That could > would instead be copied to the logical node. If a logical node exists > the RTN should have an if statement so the same code does not execute again. > > Any takers? > > Mark > > > On 19/02/2011 05:20, Simon Chen wrote: >> Hi all, >> >> I know this is kinda an old topic, but I just couldn't get it working. >> >> Here is a previous attempt using insertLogical() to handle transitive >> closure: >> http://drools-java-rules-engine.46999.n3.nabble.com/transitive-closure-td... >> The problem with this one is that the newly "logically inserted" object >> would violate its own "not exists" condition term, thus removing itself, >> then goes the infinite circle of insert/remove... >> >> Here is a post that deals with transitive closure using "insert", but it >> doesn't handle object removal correctly: >> http://drools-java-rules-engine.46999.n3.nabble.com/one-question-about-Tr... >> >> >> To me, using insertLogical is attractive because it doesn't require me to >> write specific rules to handle object removal. Is there a trick that I can >> use to actually implement transitive closure with insertLogical? >> >> Thanks a lot! >> -Simon >> _______________________________________________ >> rules-users mailing list >> rules-users(a)lists.jboss.org >> https://lists.jboss.org/mailman/listinfo/rules-users >> >> > > _______________________________________________ > rules-users mailing list > rules-users(a)lists.jboss.org > https://lists.jboss.org/mailman/listinfo/rules-users > _______________________________________________ rules-users mailing list rules-users(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/rules-users

On 19/02/2011 15:01, Simon Chen wrote: > Let me make sure I get it correctly... > > The example you gave seems to be the one-hop case. For the two-hop > case, we need something like this > > when >     edge(a, b), reach(b, c), not exists reach(a, c) > then >     insertLogical( reach(a,c) ) > > So, where do you put your logical around? It should include both > edge(a,b) and reach(b,c), right? > > Another thought, can we have something like > testExistsAndInsertLogical() to replace insertLogical()? But this may > be buggy, as the conditions are all met, so the rule actually fired... > > I really need this working, so I'll probably take a crack at modifying > drools, although haven't done it before... I'm just thinking about what it is you want to do. Do you wan to pre-compute a matrix for fast lookup if C is reachable from A? Where the integrity of that matrix is maintained by logical insertions? Or are you trying to do adhoc reachability with minimial maintained state, no pre-calculated state, where performance is not such an issue?

Mark > Thanks. > -Simon > > On Saturday, February 19, 2011, Mark Proctor<mproctor(a)codehaus.org>  wrote: >> I think to impl this what is needed is a "logical" node. At the moment >> the entire LHS forms the justification. But if we supported a logical >> node, we could do this: >> rule "reachDirect" >>       salience 10 >>       when >>           logical( e : Edge(s1 : source, t1 : target) ) >>           not( Reach(source == s1, target == t1) ) >>       then >>           insertLogical( new Reach(e.getSource(),e.getTarget()) ); >>           System.out.println( "Reach " + e.getSource() + "," + >> e.getTarget() ); >> end >> >> >> That means that it would be inserted when there was no Reach, but it >> would only be retracted when there was no matching Edge. The >> justification is only for the part of the rule that is in the logical >> grouping. >> >> To do this is actually quite a trivial change in drools, but it's not >> something we do now. I think one reason why I held off was that i was >> looking at Jess and Clips that have this and they state you can have >> multiple logical elements. But i could't figure out how having 2 or 3 >> would differ, compare to having just one. >> >> Anway to support a singe logical element, you'd need to update the >> parser to support 'logical' conditional element, in the same format as >> 'not' and 'exists'. Then if you look at RuleTerminalNode you'll see the >> part of the code that is related to removing the justifications,  on a >> retract or modify - removeLogicalDependencies. Likewise if you look in >> the DefaultKnowlegeHelper you'll see how the insertion works. That could >> would instead be copied to the logical node. If a logical node exists >> the RTN should have an if statement so the same code does not execute again. >> >> Any takers? >> >> Mark >> >> >> On 19/02/2011 05:20, Simon Chen wrote: >>> Hi all, >>> >>> I know this is kinda an old topic, but I just couldn't get it working. >>> >>> Here is a previous attempt using insertLogical() to handle transitive >>> closure: >>> http://drools-java-rules-engine.46999.n3.nabble.com/transitive-closure-td... >>> The problem with this one is that the newly "logically inserted" object >>> would violate its own "not exists" condition term, thus removing itself, >>> then goes the infinite circle of insert/remove... >>> >>> Here is a post that deals with transitive closure using "insert", but it >>> doesn't handle object removal correctly: >>> http://drools-java-rules-engine.46999.n3.nabble.com/one-question-about-Tr... >>> >>> >>> To me, using insertLogical is attractive because it doesn't require me to >>> write specific rules to handle object removal. Is there a trick that I can >>> use to actually implement transitive closure with insertLogical? >>> >>> Thanks a lot! >>> -Simon >>> _______________________________________________ >>> rules-users mailing list >>> rules-users(a)lists.jboss.org >>> https://lists.jboss.org/mailman/listinfo/rules-users >>> >>> >> >> _______________________________________________ >> rules-users mailing list >> rules-users(a)lists.jboss.org >> https://lists.jboss.org/mailman/listinfo/rules-users >> > _______________________________________________ > rules-users mailing list > rules-users(a)lists.jboss.org > https://lists.jboss.org/mailman/listinfo/rules-users > > _______________________________________________ rules-users mailing list rules-users(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/rules-users

The example you gave seems to be the one-hop case. For the two-hop case, we need something like this when edge(a, b), reach(b, c), not exists reach(a, c) then insertLogical( reach(a,c) ) So, where do you put your logical around? It should include both edge(a,b) and reach(b,c), right? Another thought, can we have something like testExistsAndInsertLogical() to replace insertLogical()? But this may be buggy, as the conditions are all met, so the rule actually fired...

On Sun, Feb 20, 2011 at 8:20 PM, Ansgar Konermann <ansgar.konermann(a)googlemail.com&gt; wrote: > On 19.02.2011 16:01, Simon Chen wrote: >> The example you gave seems to be the one-hop case. For the two-hop >> case, we need something like this >> >> when >> edge(a, b), reach(b, c), not exists reach(a, c) >> then >> insertLogical( reach(a,c) ) >> >> So, where do you put your logical around? It should include both >> edge(a,b) and reach(b,c), right? >> >> Another thought, can we have something like >> testExistsAndInsertLogical() to replace insertLogical()? But this may >> be buggy, as the conditions are all met, so the rule actually fired... >> > Hi, > > from my experience, insertLogical does exactly what > testExistsAndInsertLogical would suggest. If the same object is already > in the working memory, it keeps this object and does not insert another > instance. This behaviour is not stated explicitly in the documentation, > but I did a learning test a few weeks ago and IIRC it clearly showed > this behaviour (at least for 5.0.1). -- I consider this behaviour a > feature and would like it to be kept this way. I am using Drools 5.1.1, and I don't think insertLogical prevents duplicates automatically. This also boils down to the question of how Drools decides whether two objects are indeed the same. For strings and integers, it is straightforward, but not much so for complex objects. Is there a way to pass in a comparison function?

> With this, all which is necessary to implement transitive closure is to > remove the contradicting part of the precondition to avoid oscillation. > If it turns out that insertLogical does not perform a "does fact already > exist" check and thus might potentially insert duplicates, put exists( ) > around the two preconditions and also use "exists( reach(x,y) )" to > check whether y is reachable from x. I don't quite follow. Can you elaborate with an actual rule? Thanks! -Simon > Kind regards > > Ansgar > _______________________________________________ > rules-users mailing list > rules-users(a)lists.jboss.org > https://lists.jboss.org/mailman/listinfo/rules-users > _______________________________________________ rules-users mailing list rules-users(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/rules-users

On 21/02/2011 03:03, Simon Chen wrote: > On Sun, Feb 20, 2011 at 8:20 PM, Ansgar Konermann > <ansgar.konermann(a)googlemail.com&gt; wrote: > >> On 19.02.2011 16:01, Simon Chen wrote: >> >>> The example you gave seems to be the one-hop case. For the two-hop >>> case, we need something like this >>> >>> when >>> edge(a, b), reach(b, c), not exists reach(a, c) >>> then >>> insertLogical( reach(a,c) ) >>> >>> So, where do you put your logical around? It should include both >>> edge(a,b) and reach(b,c), right? >>> >>> Another thought, can we have something like >>> testExistsAndInsertLogical() to replace insertLogical()? But this may >>> be buggy, as the conditions are all met, so the rule actually fired... >>> >>> >> Hi, >> >> from my experience, insertLogical does exactly what >> testExistsAndInsertLogical would suggest. If the same object is already >> in the working memory, it keeps this object and does not insert another >> instance. This behaviour is not stated explicitly in the documentation, >> but I did a learning test a few weeks ago and IIRC it clearly showed >> this behaviour (at least for 5.0.1). -- I consider this behaviour a >> feature and would like it to be kept this way. >> > I am using Drools 5.1.1, and I don't think insertLogical prevents > duplicates automatically. This also boils down to the question of how > Drools decides whether two objects are indeed the same. For strings > and integers, it is straightforward, but not much so for complex > objects. Is there a way to pass in a comparison function? > InsertLogical operates on equality mode, that is determined by the pojo's equals() method implementation. If an object already exists that is equal, it will use that and the justification counter for that existing object is increased. Mark >> With this, all which is necessary to implement transitive closure is to >> remove the contradicting part of the precondition to avoid oscillation. >> If it turns out that insertLogical does not perform a "does fact already >> exist" check and thus might potentially insert duplicates, put exists( ) >> around the two preconditions and also use "exists( reach(x,y) )" to >> check whether y is reachable from x. >> > I don't quite follow. Can you elaborate with an actual rule? >

wrote:

If insertLogical did introduce duplicate objects, we'd need this: when exists edge(a, b) exists reach(b, c) then insertLogical( reach(a,c) )

Kind regards Ansgar _______________________________________________ rules-users mailing list rules-users(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/rules-users

On 21/02/2011 03:03, Simon Chen wrote: > On Sun, Feb 20, 2011 at 8:20 PM, Ansgar Konermann > <ansgar.konermann(a)googlemail.com&gt;  wrote: >> On 19.02.2011 16:01, Simon Chen wrote: >>> The example you gave seems to be the one-hop case. For the two-hop >>> case, we need something like this >>> >>> when >>>      edge(a, b), reach(b, c), not exists reach(a, c) >>> then >>>      insertLogical( reach(a,c) ) >>> >>> So, where do you put your logical around? It should include both >>> edge(a,b) and reach(b,c), right? >>> >>> Another thought, can we have something like >>> testExistsAndInsertLogical() to replace insertLogical()? But this may >>> be buggy, as the conditions are all met, so the rule actually fired... >>> >> Hi, >> >> from my experience, insertLogical does exactly what >> testExistsAndInsertLogical would suggest. If the same object is already >> in the working memory, it keeps this object and does not insert another >> instance. This behaviour is not stated explicitly in the documentation, >> but I did a learning test a few weeks ago and IIRC it clearly showed >> this behaviour (at least for 5.0.1). -- I consider this behaviour a >> feature and would like it to be kept this way. > I am using Drools 5.1.1, and I don't think insertLogical prevents > duplicates automatically. This also boils down to the question of how > Drools decides whether two objects are indeed the same. For strings > and integers, it is straightforward, but not much so for complex > objects. Is there a way to pass in a comparison function? InsertLogical operates on equality mode, that is determined by the pojo's equals() method implementation. If an object already exists that is equal, it will use that and the justification counter for that existing object is increased.

Mark >> With this, all which is necessary to implement transitive closure is to >> remove the contradicting part of the precondition to avoid oscillation. >> If it turns out that insertLogical does not perform a "does fact already >> exist" check and thus might potentially insert duplicates, put exists( ) >> around the two preconditions and also use "exists( reach(x,y) )" to >> check whether y is reachable from x. > I don't quite follow. Can you elaborate with an actual rule? > > Thanks! > -Simon > >> Kind regards >> >> Ansgar >> _______________________________________________ >> rules-users mailing list >> rules-users(a)lists.jboss.org >> https://lists.jboss.org/mailman/listinfo/rules-users >> > _______________________________________________ > rules-users mailing list > rules-users(a)lists.jboss.org > https://lists.jboss.org/mailman/listinfo/rules-users > > _______________________________________________ rules-users mailing list rules-users(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/rules-users

Implementations of equals() and hashCode() should always agree: http://www.javapractices.com/topic/TopicAction.do?Id=28

--- On Tue, 2/22/11, Simon Chen <simonchennj(a)gmail.com&gt; wrote: > From: Simon Chen <simonchennj(a)gmail.com&gt; > Subject: Re: [rules-users] unsolved myth regarding transitive closure using insertlogical... > To: "Rules Users List" <rules-users(a)lists.jboss.org&gt; > Date: Tuesday, February 22, 2011, 2:24 PM > On Sun, Feb 20, 2011 at 10:11 PM, > Mark Proctor <mproctor(a)codehaus.org&gt; > wrote: > > On 21/02/2011 03:03, Simon Chen wrote: > >> On Sun, Feb 20, 2011 at 8:20 PM, Ansgar Konermann > >> <ansgar.konermann(a)googlemail.com&gt; >  wrote: > >>> On 19.02.2011 16:01, Simon Chen wrote: > >>>> The example you gave seems to be the > one-hop case. For the two-hop > >>>> case, we need something like this > >>>> > >>>> when > >>>>      edge(a, b), reach(b, c), not > exists reach(a, c) > >>>> then > >>>>      insertLogical( reach(a,c) ) > >>>> > >>>> So, where do you put your logical around? > It should include both > >>>> edge(a,b) and reach(b,c), right? > >>>> > >>>> Another thought, can we have something > like > >>>> testExistsAndInsertLogical() to replace > insertLogical()? But this may > >>>> be buggy, as the conditions are all met, > so the rule actually fired... > >>>> > >>> Hi, > >>> > >>> from my experience, insertLogical does exactly > what > >>> testExistsAndInsertLogical would suggest. If > the same object is already > >>> in the working memory, it keeps this object > and does not insert another > >>> instance. This behaviour is not stated > explicitly in the documentation, > >>> but I did a learning test a few weeks ago and > IIRC it clearly showed > >>> this behaviour (at least for 5.0.1). -- I > consider this behaviour a > >>> feature and would like it to be kept this > way. > >> I am using Drools 5.1.1, and I don't think > insertLogical prevents > >> duplicates automatically. This also boils down to > the question of how > >> Drools decides whether two objects are indeed the > same. For strings > >> and integers, it is straightforward, but not much > so for complex > >> objects. Is there a way to pass in a comparison > function? > > InsertLogical operates on equality mode, that is > determined by the > > pojo's equals() method implementation. If an object > already exists that > > is equal, it will use that and the justification > counter for that > > existing object is increased. > > Mark, > > I did a bit debugging using the drools source code, which > is neatly > written btw. I now know what caused my problem. When > logically > inserting an object, the current implementation would > compare the > object with saved EqualityKeys in the > TruthMaintenanceSystem. The > comparison is done through hashCode(), not equals(). So, a > duplicate > (by value) object can be inserted, because the hashcode is > different. > > I am not sure if this is a design decision, but I would > rather using > equals() here. I overrode the hashCode() implementation of > my objects, > and it is working now... > > Thanks. > -Simon > > > > > Mark > >>> With this, all which is necessary to implement > transitive closure is to > >>> remove the contradicting part of the > precondition to avoid oscillation. > >>> If it turns out that insertLogical does not > perform a "does fact already > >>> exist" check and thus might potentially insert > duplicates, put exists( ) > >>> around the two preconditions and also use > "exists( reach(x,y) )" to > >>> check whether y is reachable from x. > >> I don't quite follow. Can you elaborate with an > actual rule? > >> > >> Thanks! > >> -Simon > >> > >>> Kind regards > >>> > >>> Ansgar > >>> > _______________________________________________ > >>> rules-users mailing list > >>> rules-users(a)lists.jboss.org > >>> https://lists.jboss.org/mailman/listinfo/rules-users > >>> > >> _______________________________________________ > >> rules-users mailing list > >> rules-users(a)lists.jboss.org > >> https://lists.jboss.org/mailman/listinfo/rules-users > >> > >> > > > > > > _______________________________________________ > > rules-users mailing list > > rules-users(a)lists.jboss.org > > https://lists.jboss.org/mailman/listinfo/rules-users > > > > _______________________________________________ > rules-users mailing list > rules-users(a)lists.jboss.org > https://lists.jboss.org/mailman/listinfo/rules-users > _______________________________________________ rules-users mailing list rules-users(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/rules-users