Neural-Dialogue-Generation icon indicating copy to clipboard operation
Neural-Dialogue-Generation copied to clipboard

Published 20 hours ago • verified publisher icon jiweil

backward and forward data are actually the same?

Open cgsdfc opened this issue 5 years ago • 0 comments

I followed the instructions in REAME and downloaded the OpenSubData from link and unpacked to 8.9GB and I though everything was set. I was wrong. An accident run of head -n 1 on a pair of files -- s_given_t_ and t_given_s_ told that outputs were the same. But they are supposed to have inverse target and source! So I run cmp on each pair of files and the results were: every forward file and backward file are the same.

I think this is a bug. If not purely human errors. If the dataset went wrong, what else can be right?

cgsdfc avatar Apr 12 '19 05:04 cgsdfc