the input of decoder is an one-hot vector

[bestbzw]

hi Why the input of the decoder is an one-hot vector? We usually use dense vectors(through embedding lookup function) as the input in seq2seq models.

[JACOBIN-SCTCS]

I think this is used in generating the string for where clause . It uses a pointer network in which at each time step a particular token of question is pointed out

The one hot vector would have same length as that of question with 1 corresponding to the position of current token .. the process is repeated until end tokens index gets 1

This what I have understood. Correct me if I am wrong

[JACOBIN-SCTCS]

@bestbzw you can look on paper related to pointer network

[bestbzw]

@JACOBIN-SCTCS I've read the paper"pointer network", but it doesn't mention whether it uses one-hot vectors.

I found that the Pointer-Generator-Network (Get To The Point: Summarization with Pointer-Generator Networks) uses dense vectors as the inputs of decoder, so I guess pointer network also uses dense vectors.

[JACOBIN-SCTCS]

@JACOBIN-SCTCS I've read the paper"pointer network", but it doesn't mention whether it uses one-hot vectors.

I found that the Pointer-Generator-Network (Get To The Point: Summarization with Pointer-Generator Networks) uses dense vectors as the inputs of decoder, so I guess pointer network also uses dense vectors.

Ok thank you for your info