Closed bebag closed 6 years ago
Hm, looks like the labels (query word tokens/passage word tokens) in the figure for that paper are wrong. They should be flipped.
For an accurate description of our model, please see our paper -- https://arxiv.org/abs/1704.00051.
In drqa/reader/rnn_reader.py , doc_rnn takes input_size = doc_input_size (word emb + question emb + manual features) ... and question_rnn takes input_size= args.embedding_dim
The naming is a bit confusing ... I would have called question_rnn the Stacked Bi-directional RNNs taking as input word emb + question emb + manual features .... and doc_rnn the one with word emb. That would match Figure 5 of https://openreview.net/pdf?id=HJRV1ZZAW