Menu
▾
▴
Re: [DL-Learner discussion] DL-Learner solution not including object properties
|
From: Sarker, Md K. <sar...@wr...> - 2017-05-03 17:01:03
|
Hi Lorenz, Thank you very much. With Regards, ------------- Md Kamruzzaman Sarker PhD Student Dept. of Computer Science and Engineering, Wright State University, Dayton, OH http://dase.cs.wright.edu/people/kamruzzaman-sarker ________________________________ From: Lorenz B. <spo...@st...> Sent: Wednesday, May 3, 2017 2:53 AM To: Sarker, Md Kamruzzaman; dl-...@li... Subject: Re: [DL-Learner discussion] DL-Learner solution not including object properties Hi Kamruzzaman, Hi Lorenz, Many thanks for your reply. I am sorry that I attached wrong ontology file here. Here I am attaching the corrected ontology file, corresponding configuration file and the corresponding solution file. I have did the same experiments and got similar result (solution not including object properties for father_3.conf problem, which is using father_3.owl ). no worries, thank you for the updated ontologies. Is it the case that object properties need to be connected to all individuals/classes if we want a solution which includes that object property? Well, there are two points you have to keep in mind for the current example: 1. DL-Learner tries to solve the learning problem such that a solution covers all positive examples while not covering any negative example. Especially the coverage of the positive examples is important, and DL-Learner is very strict here in. Indeed, this assumes a perfect world in which we always have perfect solutions - we know, that's usually not the case, therefore, we have a configuration parameter called "noisePercentage" which allows to lower the restrictions on the coverage of the positive examples. For instance, if you add the following line to the *.conf file alg.noisePercentage = 30 DL-Learner will also return Note, that in your example the concept "male" is already a perfect solution, thus, it will probably not find any concept with a score lower than 100% . In any case, you should always set some value for the noise as long as you don't know that the data is perfect and there must be a perfect solution. Hope this helps, but feel free to ask if you have any further questions. Cheers, Lorenz With Regards, ------------- Md Kamruzzaman Sarker PhD Student Dept. of Computer Science and Engineering, Wright State University, Dayton, OH http://dase.cs.wright.edu/people/kamruzzaman-sarker ________________________________ From: Lorenz B. <spo...@st...><mailto:spo...@st...> Sent: Tuesday, May 2, 2017 5:38 AM To: dl-...@li...<mailto:dl-...@li...> Subject: Re: [DL-Learner discussion] DL-Learner solution not including object properties Hi, unfortunately your ontologies do not contain what you say, i.e. * father_2 is the same as father_0 and does not contain an additional class "transgender" * father_3 contains a class "transgender" but not the classes "human" and "animal" and furthermore not the class assertions human(zaman), animal(cow) This is especially problematic since your conf file states that there is a positive example "zaman". Note, "male" is already a trivial solution for your configuration of the learning problem file, thus, it's not that clear which other solution you expect. Cheers, Lorenz Dear Concern, With due respect, I am exploring DL-Learner and running it for Learning Problem (Class learning). >From the experiments I am doing, I am not getting the solution which includes objectProperties. I am doing the experiments using the default father.conf (which using father.owl file). When using the default father.owl file the result includes hasChild object properties. But if I include a couple of extra classes in ontology and run with the same configuration file then the result does not include object properties. Does it mean, If we need a solution including object properties, is it required for every owlClass have to be related to the object property? The experiments result I have attached here, I have attached here the configuration and ontology file here for reference. ##Original ontology given in example folder of DL-Learner. I renamed it with father_0.owl solutions: 1: male (pred. acc.: 100.00%, F-measure: 100.00%) 2: male (pred. acc.: 100.00%, F-measure: 100.00%) 3: not (female) (pred. acc.: 100.00%, F-measure: 100.00%) 4: male or (hasChild some female) (pred. acc.: 100.00%, F-measure: 100.00%) 5: (not (female)) or (hasChild some female) (pred. acc.: 100.00%, F-measure: 100.00%) 6: male or (hasChild some (not (male))) (pred. acc.: 100.00%, F-measure: 100.00%) 7: (not (female)) or (hasChild some (not (male))) (pred. acc.: 100.00%, F-measure: 100.00%) 8: male or (hasChild some (hasChild some Thing)) (pred. acc.: 100.00%, F-measure: 100.00%) 9: male or (hasChild some (hasChild some male)) (pred. acc.: 100.00%, F-measure: 100.00%) 10: male or (hasChild some (hasChild some female)) (pred. acc.: 100.00%, F-measure: 100.00%) ##father_2.owl ##original ontology with transgender class and individual boby solutions: 1: male (pred. acc.: 100.00%, F-measure: 100.00%) 2: not (female) (pred. acc.: 100.00%, F-measure: 100.00%) 3: male or transgender (pred. acc.: 100.00%, F-measure: 100.00%) 4: transgender or (not (female)) (pred. acc.: 100.00%, F-measure: 100.00%) 5: male and (not (transgender)) (pred. acc.: 100.00%, F-measure: 100.00%) 6: male or (hasChild some transgender) (pred. acc.: 100.00%, F-measure: 100.00%) 7: male or (hasChild some female) (pred. acc.: 100.00%, F-measure: 100.00%) 8: (not (female)) and (not (transgender)) (pred. acc.: 100.00%, F-measure: 100.00%) 9: (not (female)) or (hasChild some female) (pred. acc.: 100.00%, F-measure: 100.00%) 10: male or (hasChild some (not (male))) (pred. acc.: 100.00%, F-measure: 100.00%) ##father_3.owl ##original ontology with transgender, human and animal class ## individual human(zaman) , individual animal(cow) solutions: 1: male (pred. acc.: 100.00%, F-measure: 100.00%) 2: not (female) (pred. acc.: 100.00%, F-measure: 100.00%) 3: male or transgender (pred. acc.: 100.00%, F-measure: 100.00%) 4: human or male (pred. acc.: 100.00%, F-measure: 100.00%) 5: animal or male (pred. acc.: 100.00%, F-measure: 100.00%) 6: transgender or (not (female)) (pred. acc.: 100.00%, F-measure: 100.00%) 7: human or (not (female)) (pred. acc.: 100.00%, F-measure: 100.00%) 8: animal or (not (female)) (pred. acc.: 100.00%, F-measure: 100.00%) 9: male and (not (transgender)) (pred. acc.: 100.00%, F-measure: 100.00%) 10: male and (not (human)) (pred. acc.: 100.00%, F-measure: 100.00%) ------------- Md Kamruzzaman Sarker PhD Student Dept. of Computer Science and Engineering, Wright State University, Dayton, OH http://dase.cs.wright.edu/people/kamruzzaman-sarker ------------------------------------------------------------------------------ Check out the vibrant tech community on one of the world's most engaging tech sites, Slashdot.org! http://sdm.link/slashdot<https://urldefense.proofpoint.com/v2/url?u=http-3A__sdm.link_slashdot&d=DwMF-g&c=3buyMx9JlH1z22L_G5pM28wz_Ru6WjhVHwo-vpeS0Gk&r=gOKJ7GGIYSN5ZikR0Wci79-a9BTFdST29ynNAKTmkas&m=Rn6HNVx7QPR-sQg2zKEIrFVRvipjJvnfUpNHQHN7mjg&s=XXVJ4EP4pm7Aks2TiPUVeJ2l_AGeR3crohUski1ZqKk&e=> _______________________________________________ dl-learner-discussion mailing list dl-...@li...<mailto:dl-...@li...> https://lists.sourceforge.net/lists/listinfo/dl-learner-discussion<https://urldefense.proofpoint.com/v2/url?u=https-3A__lists.sourceforge.net_lists_listinfo_dl-2Dlearner-2Ddiscussion&d=DwMF-g&c=3buyMx9JlH1z22L_G5pM28wz_Ru6WjhVHwo-vpeS0Gk&r=gOKJ7GGIYSN5ZikR0Wci79-a9BTFdST29ynNAKTmkas&m=Rn6HNVx7QPR-sQg2zKEIrFVRvipjJvnfUpNHQHN7mjg&s=fE8SOJoa1PDfwXIf6j6_kaDDaw9RvDG_wevcTY3Rg6g&e=> |
