[rules-users] unsolved myth regarding transitive closure using insertlogical...

[Mark Proctor]

On 20/02/2011 17:08, Simon Chen wrote:
> On Sun, Feb 20, 2011 at 2:10 AM, Mark Proctor<mproctor at codehaus.org>  wrote:
>> 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?
Doesn't quite answer my question.

It is possible to create a pre-computed reachability matrix, that gives 
constant time lookup for reachability. You will produce an object for 
every possible query request, so it'll use a lot of memory - but if 
performance is necessary, then the trade of is worth it. I think this is 
actually the easier solution to implement too.

the other approach would be more of an adhoc query that uses more 
minimal state but has linear search times based on the eventual search 
space, as it will do an exhaustive search each time.

Mark
> Hi Mark,
>
> Well, I came from the datalog world, where such issue doesn't exist.
> Basically, I am trying to mimic datalog's reachability rules in
> Drools. In datalog, we simply put:
> reachable(a,b) :- link(a,b)
> reachable(a,b) :- link(a,z), reachable(z,b)
>
> Insertion is always easy to handle, but deletion isn't, unless we keep
> some sort of dependency among different objects. It seems that
> insertLogic() in Drools is an easy way out, except for the tricky
> issue I mentioned in the original post.
>
> To answer your question, I have all sort of different objects flying
> around, and want to derive certain knowledge based on those objects.
> The derived knowledge should be dependent on the existing objects,
> which may change over time. This reachability is just an example of
> such recursive dependency.
>
> Thanks.
> -Simon
>
>> Mark
>>> Thanks.
>>> -Simon
>>>
>>> On Saturday, February 19, 2011, Mark Proctor<mproctor at 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-td56855.html#a56858
>>>>> 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-Transitive-Closure-td57289.html
>>>>>
>>>>>
>>>>> 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 at lists.jboss.org
>>>>> https://lists.jboss.org/mailman/listinfo/rules-users
>>>>>
>>>>>
>>>> _______________________________________________
>>>> rules-users mailing list
>>>> rules-users at lists.jboss.org
>>>> https://lists.jboss.org/mailman/listinfo/rules-users
>>>>
>>> _______________________________________________
>>> rules-users mailing list
>>> rules-users at lists.jboss.org
>>> https://lists.jboss.org/mailman/listinfo/rules-users
>>>
>>>
>>
>> _______________________________________________
>> rules-users mailing list
>> rules-users at lists.jboss.org
>> https://lists.jboss.org/mailman/listinfo/rules-users
>>
>