search

hengyuan-hu / bottom-up-attention-vqa

An efficient PyTorch implementation of the winning entry of the 2017 VQA Challenge.
GNU General Public License v3.0
750 stars 182 forks source link

There is no '<unk>' in the dictionary.pkl #37

Open BasselAli1 opened 5 years ago

BasselAli1 commented 5 years ago

I know that there is supposed to be '<unk>' in the vocabulary so that in testing or in evaluation using validation data, the '<unk>' token will be mapped to a vector of zeros(in the glove vectors). But I don't see that in this project. Am I right? I am still new in deep learning so maybe i am wrong :sweat_smile:
screenshot from 2018-11-21 21-46-30 in the create_dictionary() function you are using all the datasets to build the vocabulary. Shouldn't we only use the train dataset? Please correct me if I am wrong :sweat_smile:

serena-li commented 1 year ago

I know that there is supposed to be '<unk>' in the vocabulary so that in testing or in evaluation using validation data, the '<unk>' token will be mapped to a vector of zeros(in the glove vectors). But I don't see that in this project. Am I right? I am still new in deep learning so maybe i am wrong 😅 screenshot from 2018-11-21 21-46-30 in the create_dictionary() function you are using all the datasets to build the vocabulary. Shouldn't we only use the train dataset? Please correct me if I am wrong 😅

I also encountered the same problem, have you solved it?