search

obryanlouis / qa

TensorFlow Models for the Stanford Question Answering Dataset
73 stars 30 forks source link

baseline_model #11

Closed benz871 closed 6 years ago

benz871 commented 6 years ago

What can be the baseline model for the squad assuming that you get the hidden states of the context and question using a bidirectional GRU separately over context and questions? Also did you one-hot encode the answer start or end? If we don't one hot encode and send them directly what could be the side effects of that?

obryanlouis commented 6 years ago

The model you suggest in the first question seems similar to FastQa to me. https://arxiv.org/pdf/1703.04816.pdf. The answer starts/ends aren't one-hot encoded in the data. There is a function in tensorflow that will do it, if you need that. I'm not sure what you mean in the last question.