Say I am using pretrained GloVe model:
TEXT.build_vocab(train, vectors=GloVe(name='6B', dim=100))
If there is a token in the test dataset that is not in the train dataset, but the token is actually in GloVe(name=‘6B’, dim=100). Torchtext will return 0 as index, instead of the actual index in GloVe(name=‘6B’, dim=100).
Is this a bug or do I miss anything?