Embedding with vocab size out of bounds about t2vec HOT 4 CLOSED

Technically, the indices is bound by the vocabulary size and it falls into [0, vocab_size), it does not make sense for the appearance of 18864.

from t2vec.

The vocab_size of the preprocess ouputs is 18866, I am wondering why you pass 18864 as the vocab_size in parameter of the network?

When V and D matrixes are 18866 *k demension, it is possible for some vocab_id to select their k-nearest cell which id is 18864 & 18865.

from t2vec.

The vocab_size should be the output of your preprocessing step. That value might be slightly different depending on your Julia version, previously the code was written in Julia 0.6 which yields 18864.

from t2vec.

Ok I got it, thanks for your patient reply!

from t2vec.