Re: [DL-Learner discussion] Questions on DL-Learner

SourceForge Headquarters 1320 Columbia Street Suite 310 San Diego, CA 92101 +1 (858) 422-6466

Hi Lorenz,

Thanks for your answer, you were right, it was not the latest version.
I tried the latest version on my "father_test" example. I got that:

------------------------------------------------------
DL-Learner command line interface
Initializing Component "OWL File"... OK (0ms)
Initializing Component "fast instance checker"... OK (490ms)
Initializing Component "PosNegLPStandard"... OK (0ms)
Initializing Component "CELOE"... OK (20ms)
Initializing Component "PCELOE"... OK (0ms)
Running algorithm instance "alg1"(CELOE)
more accurate (50,00%) class expression found: Thing
more accurate (83,33%) class expression found: male
Algorithm terminated successfully (time: 10s 0ms, 373537 descriptions 
tested, 22
1907 nodes in the search tree).

number of retrievals: 6
retrieval reasoning time: 0ms ( 0ms per retrieval)
number of instance checks: 4145956 (0 multiple)
instance check reasoning time: 452ms ( 0ms per instance check)
(complex) subsumption checks: 266 (0 multiple)
subsumption reasoning time: 96ms ( 0ms per subsumption check)
overall reasoning time: 549ms

solutions:
1: male (pred. acc.: 83,33%, F-measure: 85,71%)
2: (not female) (pred. acc.: 83,33%, F-measure: 85,71%)
3: Thing (pred. acc.: 50,00%, F-measure: 66,67%)
4: (female or male) (pred. acc.: 50,00%, F-measure: 66,67%)
5: (male or (not male)) (pred. acc.: 50,00%, F-measure: 66,67%)
6: (female or (not female)) (pred. acc.: 50,00%, F-measure: 66,67%)
7: ((not female) or (not male)) (pred. acc.: 50,00%, F-measure: 66,67%)

Running algorithm instance "alg2"(PCELOE)
more accurate (50,00%) class expression found: Thing
more accurate (83,33%) class expression found: male
Algorithm terminated successfully (time: 10s 52ms, 516119 descriptions 
tested, 3
28678 nodes in the search tree).

number of retrievals: 12
retrieval reasoning time: 0ms ( 0ms per retrieval)
number of instance checks: 10208696 (0 multiple)
instance check reasoning time: 1s 111ms ( 0ms per instance check)
(complex) subsumption checks: 552 (0 multiple)
subsumption reasoning time: 163ms ( 0ms per subsumption check)
overall reasoning time: 1s 274ms

solutions:
1: male (pred. acc.: 83,33%, F-measure: 85,71%)
2: (not female) (pred. acc.: 83,33%, F-measure: 85,71%)
3: Thing (pred. acc.: 50,00%, F-measure: 66,67%)
4: (female or male) (pred. acc.: 50,00%, F-measure: 66,67%)
5: (male or (not male)) (pred. acc.: 50,00%, F-measure: 66,67%)
6: (female or (not female)) (pred. acc.: 50,00%, F-measure: 66,67%)
7: ((not female) or (not male)) (pred. acc.: 50,00%, F-measure: 66,67%)
-------------------------------------------------------------

So, there is no solution with something like "hasAge integer[>=28]". 
Maybe, I did not use the correct parameters. I used the same file than 
"father.conf" (I just changed ks.fileName).

Regards,
Céline.

Le 12.04.2014 11:51, Lorenz Bühmann a écrit :
> Hi,
>
>  which version if DL-Learner do you use? The latest version online is
> 
> http://sourceforge.net/projects/dl-learner/files/DL-Learner/dllearner-1.0-beta-2.tar.gz/download
> [7] but we plan to upload a new one as there are many new features 
> and
> bugfixes in the current SVN version.
>  Form the error message, I assume you use a quite old version. Can 
> you
> try the latest version?
>
>  According to the modified example:
>
>  You have as positive examples:
>  stefan -> male, 28
>  markus -> male, 50
>  martin -> male, 34
>  and negative examples:
>  heinz -> male, 17
>  anna -> female, 10
>  michelle -> female, 4
>
>  So I guess you want to learn something like
>  "male AND hasAge integer[>=34]" ?
>
>  I'll check if this works, but it should be possible.
>
>  Regards,
>  Lorenz
>
> On 04/10/2014 10:53 AM, alec wrote:
>
>> Hi Lorenz,
>>
>> Thank you very much for your answers.
>> I'm planning to use DL-Learner to learn concept definitions from an
>> ontology of holiday destinations (I don't have the ontology yet). I
>> want to make sure it is possible to get definitions with
>> inferiority/superiority signs (about numerical datatype properties
>> not about cardinality restrictions).
>> For example, I would like to get something like that:
>> "Definition of a destination which is hot in Winter:
>> hasJanuaryTemperature x and
>> hasFebruaryTemperature y and
>> hasMarchTemperature z and
>> x>20 and
>> y>20 and
>> z>20".
>>
>> I tried to modify the "father.owl" file (see attachments) in
>> DL-Learner examples. I put a "hasAge" datatype property and I
>> deleted "hasChild". I was hoping to see if I could get a definition
>> with a superiority/inferiority sign about age. I got that:
>>
>> DL-Learner 2010-08-07 command line interface
>> starting component manager ... OK (82ms)
>> initialising component "OWL file" ... OK (0ms)
>> initialising component "fast instance checker" ... OK (388ms)
>> initialising component "pos neg learning problem" ... OK (0ms)
>> initialising component "OCEL" ... OK (14ms)
>>
>> starting top down refinement with: Thing (50% accuracy)
>> more accurate (83,33%) class expression found: male
>> Exception in thread "main" java.lang.OutOfMemoryError: GC overhead
>> limit exceeded
>> at java.util.LinkedList.linkLast(Unknown Source)
>> at java.util.LinkedList.add(Unknown Source)
>> at java.util.LinkedList.clone(Unknown Source)
>> at
>>
> 
> org.dllearner.refinementoperators.RhoDRDown.refine(RhoDRDown.java:474)
>>
>> at
>>
> 
> org.dllearner.refinementoperators.RhoDRDown.refine(RhoDRDown.java:498)
>>
>> at
>>
> 
> org.dllearner.refinementoperators.RhoDRDown.refine(RhoDRDown.java:470)
>>
>> at
>>
> 
> org.dllearner.refinementoperators.RhoDRDown.refine(RhoDRDown.java:413)
>>
>> at
>>
> 
> org.dllearner.algorithms.refinement2.ROLearner2.extendNodeProper(ROLearner2.java:551)
>> at
>>
> 
> org.dllearner.algorithms.refinement2.ROLearner2.extendNodeProper(ROLearner2.java:521)
>> at
>>
> 
> org.dllearner.algorithms.refinement2.ROLearner2.start(ROLearner2.java:436)
>> at
>>
> 
> org.dllearner.algorithms.refinement2.ROLComponent2.start(ROLComponent2.java:441)
>> at org.dllearner.cli.Start.start(Start.java:347)
>> at org.dllearner.cli.Start.main(Start.java:209)
>>
>> Kind regards,
>> Céline
>>
>> Le 10.04.2014 00:02, Lorenz Bühmann a écrit :
>>
>>> Hi Céline,
>>>
>>> of course we can give you more information about DL-Learner if
>>> you're
>>> interested in.
>>>
>>> 1.) I'm not exactly sure what you mean by target language, but if
>>> if you
>>> refer to what's the expressivity of the learned class
>>> expressions, then
>>> no, the target language of DL-Learner is not ALC.
>>> Depending on the used learning algorithm, DL-Learner of course
>>> supports
>>> datatype properties and for example can also learn class
>>> expressions
>>> which consist of constructs used in Description Logics beyond
>>> ALC, like
>>> for example qualified cardinality restrictions(Q).
>>>
>>> 2.) see 1.)
>>>
>>> 3.) We do not have any numbers, but in general the internally
>>> used OWL
>>> reasoner(e.g. Pellet or HermiT) might be a bottleneck. If you're
>>> able to
>>> just load the necessary part of the ontology, this can of course
>>> positively influence the learning process. Maybe we're both
>>> taking about
>>> different things when using the term "noise", but I wouldn't
>>> declare
>>> unnecessary information as noise.
>>>
>>> 4.) Limits in which sense?
>>>
>>> Can you give us any insights into what you're planning to do with
>>> the
>>> DL-Learner?
>>>
>>> Kind regards,
>>> Lorenz
>>> On 04/08/2014 09:00 AM, alec wrote:
>>>
>>>> Hello,
>>>>
>>>> I am a PhD student in Laboratoire de Recherche en Informatique
>>>> in
>>>> Université Paris Sud (France). I have read papers on
>>>> DL-Learner. For my
>>>> thesis project, I might be interested in using an ILP tool to
>>>> learn
>>>> concept definitions. But the ontology I will use as input will
>>>> have
>>>> datatype properties (numerical values) and I would like to use
>>>> a tool
>>>> which can learn
>>>> definitions using these datatype properties.
>>>>
>>>> I would like to have some additional information on DL-Learner
>>>> if it is
>>>> possible. I would be grateful if you could answer my questions.
>>>>
>>>>
>>>> 1. I understood that the target language of your algorithm is
>>>> ALC
>>>> description logic. Can you confirm me that we cannot get a
>>>> definition of
>>>> a concept with datatype properties (other than string datatype
>>>> properties)?
>>>> For example, something like an adult is a person whose age
>>>> hasValue x
>>>> with x>=18.
>>>>
>>>> 2. If I understood right:
>>>> Is there a particular reason for that? Has it a real complexity
>>>> to
>>>> implement? Or do you know tools (open source or free of charge
>>>> for
>>>> academic research) that can generate a definition with
>>>> numerical
>>>> datatype properties (e.g. in SHOIN(D) description logic)?
>>>>
>>>> 3. Are there any constraints about the input ontology? Can it
>>>> be a big
>>>> ontology with potential information which is not interesting
>>>> for
>>>> defining a concept (i.e. with noise)? Or has it to be just the
>>>> interesting part of the ontology?
>>>>
>>>> 4. Can you say what the limits of DL-Learner are?
>>>>
>>>> I would greatly appreciate any help you might be able to give
>>>> me.
>>>>
>>>> Best regards,
>>>> Céline Alec
>>>>
>>>>
>>>
>>
> 
> ------------------------------------------------------------------------------
>>>>
>>>> Put Bad Developers to Shame
>>>> Dominate Development with Jenkins Continuous Integration
>>>> Continuously Automate Build, Test & Deployment
>>>> Start a new project now. Try Jenkins in the cloud.
>>>> http://p.sf.net/sfu/13600_Cloudbees [1]
>>>> _______________________________________________
>>>> dl-learner-discussion mailing list
>>>> dl-...@li... [2]
>>>>
>>> https://lists.sourceforge.net/lists/listinfo/dl-learner-discussion
>>>> [3]
>>>
>>>
>>
> 
> ------------------------------------------------------------------------------
>>>
>>> Put Bad Developers to Shame
>>> Dominate Development with Jenkins Continuous Integration
>>> Continuously Automate Build, Test & Deployment
>>> Start a new project now. Try Jenkins in the cloud.
>>> http://p.sf.net/sfu/13600_Cloudbees [4]
>>> _______________________________________________
>>> dl-learner-discussion mailing list
>>> dl-...@li... [5]
>>> https://lists.sourceforge.net/lists/listinfo/dl-learner-discussion
>>> [6]
>>
>>
> 
> ------------------------------------------------------------------------------
>> Put Bad Developers to Shame
>> Dominate Development with Jenkins Continuous Integration
>> Continuously Automate Build, Test & Deployment
>> Start a new project now. Try Jenkins in the cloud.
>> http://p.sf.net/sfu/13600_Cloudbees
>>
>> _______________________________________________
>> dl-learner-discussion mailing list
>> dl-...@li...
>> https://lists.sourceforge.net/lists/listinfo/dl-learner-discussion
>
>
>
> Links:
> ------
> [1] http://p.sf.net/sfu/13600_Cloudbees
> [2] mailto:dl-...@li...
> [3] 
> https://lists.sourceforge.net/lists/listinfo/dl-learner-discussion
> [4] http://p.sf.net/sfu/13600_Cloudbees
> [5] mailto:dl-...@li...
> [6] 
> https://lists.sourceforge.net/lists/listinfo/dl-learner-discussion
> [7]
> 
> http://sourceforge.net/projects/dl-learner/files/DL-Learner/dllearner-1.0-beta-2.tar.gz/download