question about config parameter "only_sentence" about disan HOT 3 CLOSED

Hi,

As you said, if "only_sentence" is set to False, the phrase-level samples will be in consideration during training, whereas if it's set to True, the only sentence-level samples will be used during the training. Under any circumstance, the evaluation only is applied to the sentence during test phrase.

from disan.

Hi, @taoshen58

Thank you for your reply.

Could you tell me which setting is do you use when doing experiments ? Is it the default one: False ?

from disan.

Yes, set the 'only_sentence' to false. And the evaluation provided in this paper is only the "fine-grained" and "root" shown in Table 1 of RNTN.

And, it is also confusing to me that, it seems they (e.g., RNTN, TreeLSTM) used the fine-grained and phrase&sentence-level label for training and made the evaluation on various settings (e.g., the four settings shown in table 1 aforementioned). Please refer to here.

Welcome to discuss!

from disan.