Re: [DL-Learner discussion] dl-learner and noise

[Lorenz B. <spo...@st...>]

Hello Céline,

there is no need to apologize. This is the official DL-Learner 
discussion list and we're always happy when we can help people in using 
DL-Learner or answer questions.

1. Yes, at least for the learning algorithm CELOE there is a parameter 
called "maxNrOfResults" which limits the number of returned solutions. 
I'm not sure if I understand the second point, do you ask for another 
option that handles all solution with the same accuracy as one solution, 
thus, you want e.g. n distinct solutions with different score?

2. The solutions are sorted by:

accuracy > length > class expression type

We use the length as second criteria because of readability and 
simplicity, as you already assumed. Indeed there might be other sorting 
priorities (probably depending on the use-case and/or dataset), but the 
main focus is on the accuracy. If you have any other in mind, let us 
know and we will think about it.


“hasCulture min 6 Culture” is not in the solution list, because it's 
follows logically from the knowledge base itself. We have an optional 
parameter that rewrites the class expressions exactly like in your 
example, but that's unfortunately not yet available in the conf files. 
I've open a feature request [1].

3. You're right. This is not very efficient and makes the results more 
confusing. I wouldn't say that this is a bug, but definitely needs to be 
avoided. I opened a ticket [2].


Kind regards,
Lorenz


[1] https://github.com/AKSW/DL-Learner/issues/2
[2] https://github.com/AKSW/DL-Learner/issues/3

> Hello Lorenz,
>
> Thank you again for your answer. I wish you a happy new year. I am sorry
> to disturb you again with some new questions…
>
> 1. Is there a parameter to display more than the 10 first best
> solutions? For example, to display all the solutions with the same
> accuracy and f-measure as the first solution.
>
>
> 2. What is the criteria for the order of the answers (besides accuracy
> and f-measure)?
> If there are several solutions with the same accuracy/f-measure, how do
> you choose what solution to put in the first position?
> I think it is related to the readability, but could you explain a bit
> more please?
>
> I am asking this because I am confused with 2 things:
>
> - If I take the “hasCulture min 6 Thing” with the parameters you have
> added: the second solution is “hasCulture min 6 (not WaterBody)” and the
> next ones are all like “hasCulture min 6 (not Concept_Name)”. I don’t
> understand why “hasCulture min 6 Culture” is not at least in the second
> position. I mean, “Culture” is more readable than “(not Concept_name)”,
> isn’t it?
>
> - Moreover, in this context, “hasCulture min 6 Culture” and “hasCulture
> min 6 Thing” are equivalent because Culture is the range of hasCulture.
> But in my opinion, it would be more adapted to get “hasCulture min 6
> Culture” before “hasCulture min 6 Thing”. I mean, I would rather say to
> an expert that a cultural destination is a destination which hasCulture
> at least “6 instances of Culture” rather than “6 instances of Thing”. Of
> course, this is only a matter of readability for an expert, since saying
> “Thing” or “Culture” does not change anything if the ontology is
> consistent.
>
>
> 3. Is there a way to avoid to get solutions with the same
> accuracy/f-measure as one solution but which are much more complex for
> no reason?
>
> For example, if I disable the “not” and “only” with the “hasCulture min
> 6 Thing” example, I get:
> 1: hasCulture min 6 Thing
> 2: Destination and hasCulture min 6 Thing
> 3: hasCulture min 6 Thing and (Destination or Weather)
> 4: hasCulture min 6 Thing and (Destination or Season)
> Etc
>
> All of these definitions are 100% correct.
> But what is the point to get the “or Weather” / “or Season” part? If I
> say “hasCulture min 6 Thing and (Destination or ConceptA or ConceptB or
> …)”, of course I can add anything with “or”, it would not change the
> accuracy but I don’t see the point.
>
>
> Thanks a lot,
> Céline
>
>
> ------------------------------------------------------------------------------
> Dive into the World of Parallel Programming! The Go Parallel Website,
> sponsored by Intel and developed in partnership with Slashdot Media, is your
> hub for all things parallel software development, from weekly thought
> leadership blogs to news, videos, case studies, tutorials and more. Take a
> look and join the conversation now. http://goparallel.sourceforge.net
> _______________________________________________
> dl-learner-discussion mailing list
> dl-...@li...
> https://lists.sourceforge.net/lists/listinfo/dl-learner-discussion