It seems that the character-level embedding is not in character-level
GeneZC opened this issue · comments
According to what the paper said, a word should be split in following way:
'mars' --> 'm', 'a', 'r', 's'
and through a bi-lstm and embedding, we yield a output.
However, I notice that in 'char_mapping' operation, a sentence is regarded as the key
Am I wrong?
Oh, I get it! It's in 'prepare_dataset' you do the operation mentioned above
Sorry for that!
not 'prepare_dataset' but 'create_dico'