search

starling-lab / BoostSRL

BoostSRL: "Boosting for Statistical Relational Learning." A gradient-boosting based approach for learning different types of SRL models.
https://starling.utdallas.edu
GNU General Public License v3.0
32 stars 21 forks source link

Question about multi-target dataset #38

Closed asd36952 closed 4 years ago

asd36952 commented 4 years ago

Dear all,

I have question about multi-target dataset.

There are datasets which have more than one relation as target. For example, (1) Cora => samebib, sametitle, sameauthor, samevenue (2) Citeseer => infield_ftitle, infield_fauthor, infield_fvenue

If we want to run the code with only one relation, e.g. only samebib in Cora (-target samebib), should I move the other evidence about sametitle, sameauthor, samevenue from train(test)_pos(neg) to train(test)_facts?

And When we run the code with onle one relation, e.g. only samebib in Cora (-target samebib), I want to know whether the model will ignore the sametitle, sameauthor, samevenue evidence in train(test)_pos(neg) file or not,

Thank you, -Seongwoo

nandhiniramanan5 commented 4 years ago

Hi Seongwoo,

Yes, you must move the evidence about sametitle, sameauthor, samevenue from train(test)_pos(neg) to train(test)_facts, when reasoning about only samebib in Cora .

asd36952 commented 4 years ago

Thank you for your response!

asd36952 commented 4 years ago

Dear all,

When I move the evidence from train_neg to train_fact, "!" expression is not work.

How can I express the negation of predicate? e.g. samebib(p1,p2)=False =>!samebib(p1,p2)

or Does the BoostSRL only support the closed world assumption?

Thank you, -Seongwoo

nandhiniramanan5 commented 4 years ago

BoostSRL makes closed world assumption. So you only have to specify the evidence that are true in the domain.

asd36952 commented 4 years ago

I understand. Thank you!