I'm not sure what you're asking for here...
Prediction needs to train a machine learning (ML) model. Training an ML
model, needs to think about relevant features first, then feed the data
into the ML algorithm, then use the model to predict things on new data.
Why can't you define the relevant features, then select them from the
data, then do the ML part?
Indeed, you could "tag" the relevant properties in your data, just add
triples about them, (.e.g :p :featureLevel 3 .). Then select only data
with properties that you need, e.g.
select ... {?s ?p ?o . ?p :featureLevel 1}
or whatever. There is no black magic behind modeling, adding, and
querying data in RDF.
On 09.08.2018 13:34, javed khan wrote:
> Hello Lorenz,
>
> I also have same thinking i.e to make (data property) values (like 1,2,3)
>> for each feature and then fetch them . For example, we have dataset of a
>> Student and want to predict their performance, with features, name,
>> midExam-score, finalExam-score, lecture-attendance, address, etc.. Then
>> certainly we dont need here features like name, address so we will assign
>> these features a different number (say 2) and assign a number 1 to the
>> features which have more relevance.
>
> Lorenz, if you could give me (based on your experience of Semantic Web)
> some suggestions here to implement this procedure.
>
> Thanks
>
> On Thu, Aug 9, 2018 at 11:16 AM, Lorenz Buehmann <
> buehm...@informatik.uni-leipzig.de> wrote:
>
>> no no no ...
>>
>> SPARQL is a query language and not a machine learning algorithm.
>>
>> RDF dataset D -> SPARQL query Q -> Q(D) = part of the data by means of
>> graph pattern matching. there is no induction, confidence values, etc.
>>
>> define "most appropriate", then write the query to select those
>> "features" - to repeate myself, your query just returns facts from the
>> RDF dataset. nothing more, nothing less. which facts has to be defined
>> by yourself.
>>
>>
>> On 09.08.2018 09:08, javed khan wrote:
>>> Hello Lorenz sorry for inconvenience.
>>>
>>> Suppose we have a dataset about software effort estimation which have
>> about
>>> 100 attributes (line of code, complexity etc). To extract the subset of
>>> features (as all 100 attributes are not important), we usually used some
>>> data mining algorithms (Genetic algorithm etc). Now I want to make these
>>> attributes as Ontology and get the attributes via SPARQL but not sure how
>>> can I select the most appropriate features using SPARQL (because the
>>> Genetic algorithm and other select the most appropriate features
>>> automatically).
>>>
>>> Best regards
>>>
>>> On Thu, Aug 9, 2018 at 7:53 AM, Lorenz Buehmann <
>>> buehm...@informatik.uni-leipzig.de> wrote:
>>>
>>>> Javed...I don't know how many mails you wrote to this list (and the
>>>> Protege Users list), but shouldn't you already have learned that you
>>>> have to provide more information, data, code, queries, use case, etc. in
>>>> order to get help?!
>>>>
>>>> What do you expect as an answer on a question "can we ..." - my answer
>>>> is, "yes, you can". And now?
>>>>
>>>> Just to clarify, SPARQL is a query language for RDF data, thus, you can
>>>> query any explicit information that is contained in an RDF dataset.
>>>>
>>>>
>>>> On 08.08.2018 21:56, javed khan wrote:
>>>>> Hello
>>>>>
>>>>> For instance, we have to select some features ( for data mining) , can
>> we
>>>>> do it using Semantic Web technologies like Ontology and SPARQL. ?
>>>>>
>>>>> Thanks
>>>>>
>>
>>
