LISTSERV mailing list manager LISTSERV 16.0

Help for ECJ-INTEREST-L Archives


ECJ-INTEREST-L Archives

ECJ-INTEREST-L Archives


ECJ-INTEREST-L@LISTSERV.GMU.EDU


View:

Message:

[

First

|

Previous

|

Next

|

Last

]

By Topic:

[

First

|

Previous

|

Next

|

Last

]

By Author:

[

First

|

Previous

|

Next

|

Last

]

Font:

Proportional Font

LISTSERV Archives

LISTSERV Archives

ECJ-INTEREST-L Home

ECJ-INTEREST-L Home

ECJ-INTEREST-L  May 2013

ECJ-INTEREST-L May 2013

Subject:

Re: Postprocessing in ec.gp

From:

Uday kamath <[log in to unmask]>

Reply-To:

ECJ Evolutionary Computation Toolkit <[log in to unmask]>

Date:

Fri, 24 May 2013 05:02:04 -0400

Content-Type:

text/plain

Parts/Attachments:

Parts/Attachments

text/plain (226 lines)

Bojan
My Rules are classification rules and have binary classification with
Attributes and Class (positive and negative)
Uday

On Fri, May 24, 2013 at 4:44 AM, Bojan Janisch
<[log in to unmask]> wrote:
> Hey Uday,
>
> thanks to you I got the first version of my GPTree (still not usable). I'm still not used with the representation of a rule as some tree constraints, but I think this is only a matter of usage.
> What do you mean with positive class and negative class?
>
> Greetings
>
> Bojan
>
> ----- Ursprüngliche Mail -----
> Von: "Uday kamath" <[log in to unmask]>
> An: [log in to unmask]
> Gesendet: Freitag, 24. Mai 2013 03:52:40
> Betreff: Re: Postprocessing in ec.gp
>
> Bojan
>
> As i wrote before that my Rules are of condition->action format. The
> action is determined by evaluating the GP Individual where fitness is
> the "rule coverage" for the class.
>
>
> The condition is a GPTree with constraints defined as below
>
> Boolean Functions: AND, OR, NOT takes condition or boolean returns boolean
> Operators: >,<,= takes two nodes attribute and value ranges and
> returns condition
> Attribute: (Set of all attribute values) leaf node with no child
> Value: ERCs or Constants, leaf node with no child
>
> Basically the GPTrees forms complex Rules like
> (Person.age > 45) AND (Person.state = Virginia)
>
> GPIndividual covers positive class in 80% cases and negative class in
> 20%, so the discrimination power of rule and coverage become the
> fitness value.
>
> Hope this helps
> Uday
>
>
> On Thu, May 23, 2013 at 9:40 AM, Bojan Janisch
> <[log in to unmask]> wrote:
>> Hey Uday,
>>
>> I appreciate your help and I'm sorry to bother you further but I don't get the GPTree working.
>> Currently I'm hanging on the problem that a GPTypeSet cannot have a GPTypeSet as it's child,
>> maybe because of the growth.
>>
>> Could you explain me with an example how you represented a rule as a GPTree?
>>
>> Greetings from Germany
>>
>> Bojan
>>
>> ----- Ursprüngliche Mail -----
>> Von: "Uday kamath" <[log in to unmask]>
>> An: [log in to unmask]
>> Gesendet: Donnerstag, 16. Mai 2013 16:45:03
>> Betreff: Re: Postprocessing in ec.gp
>>
>> Bojan
>>
>> You need to give the parameters needed, the manual specifies them and
>> here is the example
>>
>> ***************************************************************************************************
>> pop.subpop.0.species.pipe.source.0=ec.gp.breed.SizeFairCrossoverPipeline
>> gp.breed.size-fair.source.0=ec.select.TournamentSelection
>> gp.breed.size-fair.source.1=same
>> gp.breed.size-fair.ns.0=ec.gp.koza.KozaNodeSelector
>> gp.breed.size-fair.ns.1=ec.gp.koza.KozaNodeSelector
>> gp.breed.size-fair.tries=20
>> gp.breed.size-fair.homologous=true
>> gp.breed.size-fair.maxdepth=25
>> ***********************************************************************************************************
>>
>> Uday
>>
>>
>> On Thu, May 16, 2013 at 10:10 AM, Bojan Janisch
>> <[log in to unmask]> wrote:
>>> Hey Uday,
>>>
>>> I'm trying to get the SizeFairCrossoverPipeline into my BreedingPipeline, but seems that I'm missing an argument or so. I'm getting this error stack:
>>>
>>> Initializing Generation 0
>>> Exception in thread "main" ec.util.ParamClassLoadException:
>>> No class name provided.
>>> PARAMETER: pop.subpop.0.species.pipe.source.0.source.0
>>>      ALSO: gp.breed.size-fair.source.0
>>>         at ec.util.ParameterDatabase.getInstanceForParameter(ParameterDatabase.java:420)
>>>         at ec.BreedingPipeline.setup(BreedingPipeline.java:190)
>>>         at ec.gp.breed.SizeFairCrossoverPipeline.setup(SizeFairCrossoverPipeline.java:160)
>>>         at ec.BreedingPipeline.setup(BreedingPipeline.java:192)
>>>         at ec.breed.MultiBreedingPipeline.setup(MultiBreedingPipeline.java:66)
>>>         at ec.Species.setup(Species.java:218)
>>>         at ec.gp.GPSpecies.setup(GPSpecies.java:45)
>>>         at ec.Subpopulation.setup(Subpopulation.java:166)
>>>         at ec.Population.setup(Population.java:142)
>>>         at ec.simple.SimpleInitializer.setupPopulation(SimpleInitializer.java:55)
>>>         at ec.simple.SimpleInitializer.initialPopulation(SimpleInitializer.java:46)
>>>         at ec.simple.SimpleEvolutionState.startFresh(SimpleEvolutionState.java:54)
>>>         at ec.EvolutionState.run(EvolutionState.java:438)
>>>         at ec.Evolve.main(Evolve.java:750)
>>>
>>> while using these parameters for breeding pipeline:
>>>
>>> op.subpop.0.species.pipe = ec.breed.MultiBreedingPipeline
>>> # Koza's decision here was odd...
>>> pop.subpop.0.species.pipe.generate-max = false
>>> # Subsidiary pipelines:
>>> pop.subpop.0.species.pipe.num-sources = 2
>>> pop.subpop.0.species.pipe.source.0 = ec.gp.breed.SizeFairCrossoverPipeline
>>> pop.subpop.0.species.pipe.source.0.prob = 0.9
>>> pop.subpop.0.species.pipe.source.1 = ec.breed.ReproductionPipeline
>>> pop.subpop.0.species.pipe.source.1.prob = 0.1
>>>
>>> Simply changed the CrossoverPipeline to SizeFairCrossoverPipeline in the koza.params which I'm using as parent params-file. Could you please also explain
>>> how do you managed to get a variable amount of conditions for your rule-tree? Or does every rule has the same amount of conditions?
>>> Currently I'm using 5 different tree constraints where each tree represent a rule with a different amount of children. (I don't know set this to random,
>>> maybe if I know more about the evolution process in detail, but I don't know where I could find such an information)
>>>
>>> Greetings from Germany
>>>
>>> Bojan
>>>
>>> ----- Ursprüngliche Mail -----
>>> Von: "Uday kamath" <[log in to unmask]>
>>> An: [log in to unmask]
>>> Gesendet: Freitag, 26. April 2013 01:14:26
>>> Betreff: Re: AW: Re: Postprocessing in ec.gp
>>>
>>>
>>> Bojan
>>>
>>>
>>> Some of the measures i have in my Rules generation are
>>> 1. Use Homologous Crossover instead of Standard Subtree Crossover, works very effectively in reducing the size or 'bloat' of trees. ECJ now has the Homologous or SizeFair crossover.
>>> 2. As sean mentioned, i have mutation operator that "removes" the subtree in the pipeline. Only thing is if the fitness is not changed or below an epsilon, i accept it. ECJ has bunch of delete or raise mutation for GP trees.
>>> 3. In my final interpreter i have a Simplifier for Boolean Expression, which does logical reduction based on karnaugh maps.
>>>
>>>
>>> With these three mechanism i was able to get meaningful and pruned Rules.
>>> Hope that helps
>>> Uday
>>>
>>>
>>>
>>> On Thu, Apr 25, 2013 at 6:58 PM, Sean Luke < [log in to unmask] > wrote:
>>>
>>>
>>> Pruning will be up to you. But you have two ways you can hook into ECJ to do it.
>>>
>>> 1. You can create a BreedingPipeline which prunes the trees handed it by its children. This is essentially a mutation mechanism during breeding of individuals. You'd probably set this up as the top-level pipeline, with the standard Koza pipeline structure underneath it, so it's the last thing that operates on new individuals.
>>>
>>> 2. Perhaps a better approach would be simply to prune the individual during fitness evaluation prior to testing it.
>>>
>>> Sean
>>>
>>> Begin forwarded message:
>>>
>>>> From: Bojan Janisch < [log in to unmask] >
>>>> Date: April 25, 2013 5:27:32 PM EDT
>>>> To: [log in to unmask]
>>>> Subject: AW: Re: Postprocessing in ec.gp
>>>> Reply-To: ECJ Evolutionary Computation Toolkit < [log in to unmask] >
>>>
>>>
>>>>
>>>> Hi Uday,
>>>> I already have a plan how the represented rule should be evaluated. I'll write an
>>>> API to JBoss Drools, which is a rule-based System. The Output of the System will
>>>> be compared to a goldstandard by a comparator which I have to write. It returns a
>>>> F-Score or something similar to it. The Score goes into the fitness function.
>>>>
>>>> But my problem is the pruning. I need a step where I can check a syntax tree for useless
>>>> part-trees. Otherwise the trees have no limits on how big it can grow. I wanted to check
>>>> while post-processing the tree, but seems that ECJ does not support it.
>>>>
>>>> How did you solve the chunk-condition problem for rules?
>>>>
>>>> Greetings
>>>> Bojan
>>>>
>>>>
>>>> ----- Ursprüngliche Mail -----
>>>> Von: Uday kamath < [log in to unmask] >
>>>> An: [log in to unmask]
>>>> Gesendet: Thu, 25 Apr 2013 17:28:38 +0200 (CEST)
>>>> Betreff: Re: Postprocessing in ec.gp
>>>>
>>>> Bojan
>>>> There are no pre written post processing out of the box as it depends on
>>>> the problem (function set of GP). I will give a short background which you
>>>> may find useful for what you are doing.
>>>> For generating automated Rules to discriminate certain patterns, i had my
>>>> GP have basic terminals and non-terminals required for the Rule. I would
>>>> collect my "exceptional" Rules as a memory pad outside of evolutionary
>>>> runs, everytime i see some well performing ones. Finally i get a Rule set
>>>> and wrote my Custom "interpreter" that can parse the GP Individuals as
>>>> s-expression to evaluate. So ECJ gives you GP and Evolution for free, you
>>>> need to write custom functions and interpreter for them.
>>>> Hope that helps
>>>> Uday
>>>>
>>>>
>>>> On Thu, Apr 25, 2013 at 10:29 AM, Bojan Janisch <
>>>> [log in to unmask] > wrote:
>>>>
>>>>> Hello everyone,
>>>>>
>>>>> is there a post-processing-step in the Genetic Programming algorithm? I've
>>>>> search through the manual,
>>>>> but did not find an explanation about this topic.
>>>>>
>>>>> Thank you and greetings,
>>>>> Bojan
>>>>>

Top of Message | Previous Page | Permalink

Advanced Options


Options

Log In

Log In

Get Password

Get Password


Search Archives

Search Archives


Subscribe or Unsubscribe

Subscribe or Unsubscribe


Archives

September 2019
August 2019
June 2019
April 2019
March 2019
January 2019
December 2018
November 2018
October 2018
July 2018
May 2018
January 2018
December 2017
November 2017
October 2017
September 2017
August 2017
July 2017
June 2017
May 2017
April 2017
March 2017
February 2017
January 2017
December 2016
November 2016
October 2016
September 2016
August 2016
July 2016
June 2016
May 2016
April 2016
March 2016
February 2016
January 2016
December 2015
November 2015
October 2015
September 2015
August 2015
July 2015
June 2015
May 2015
April 2015
March 2015
February 2015
January 2015
December 2014
November 2014
September 2014
August 2014
July 2014
June 2014
May 2014
April 2014
March 2014
February 2014
January 2014
December 2013
November 2013
October 2013
September 2013
August 2013
July 2013
June 2013
May 2013
April 2013
March 2013
February 2013
January 2013
December 2012
November 2012
October 2012
September 2012
July 2012
June 2012
May 2012
April 2012
March 2012
February 2012
January 2012
December 2011
November 2011
October 2011
September 2011
August 2011
July 2011
June 2011
May 2011
April 2011
March 2011
February 2011
January 2011
December 2010
November 2010
October 2010
September 2010
August 2010
July 2010
June 2010
May 2010
April 2010
March 2010
February 2010
January 2010
December 2009
November 2009
October 2009
September 2009
August 2009
July 2009
June 2009
May 2009
April 2009
March 2009
February 2009
January 2009
December 2008
November 2008
October 2008
September 2008
August 2008
July 2008
June 2008
May 2008
April 2008
March 2008
February 2008
January 2008
December 2007
November 2007
October 2007
September 2007
August 2007
July 2007
June 2007
May 2007
April 2007
March 2007
February 2007
January 2007
December 2006
November 2006
October 2006
September 2006
August 2006
July 2006
June 2006
May 2006
April 2006
March 2006
February 2006
January 2006
December 2005
November 2005
October 2005
September 2005
August 2005
July 2005
June 2005
May 2005
April 2005
March 2005
February 2005
January 2005
December 2004
November 2004
September 2004
August 2004
July 2004
June 2004
May 2004
April 2004
March 2004

ATOM RSS1 RSS2



LISTSERV.GMU.EDU

CataList Email List Search Powered by the LISTSERV Email List Manager