Is it a bug to increase the best_ind in decoder_rnn.py?

[fengyang0317]

trafficstars

https://github.com/rowanz/neural-motifs/blob/master/lib/lstm/decoder_rnn.py#L227 It seems that the best_ind is already in the 1-150. Why do you add best_ind by 1 here?

[rowanz]

hi there,

good question! looking at it again, I'm not sure what I was doing there. unfortunately I don't have too much time to investigate this at the moment, but it's possible there was a bug in the code. Note that this was the code responsible for all the experiments in the paper, so fixing this might increase performance - if it was indeed a problem.

curious to hear your thoughts!!

Rowan

[simonJJJ]

Hi, I think in decoder_rnn.py line56, Rowan had added a ['start'] in the zero index. So is it actually 2-151 ?