Reproducibility of the results

[594zyc]

Hi,

Thanks for providing the source code!
I tried running the code but obtained some weird results like this

seed	bleu	match	success_f1	info_f1	req_f1
0	0.207949275	(0.3880597014635776, 0.0)	(0.7801418389929059, 0.8536585365664378, 0.7182835820761514)	0.722941476	0.87251
1	0.200348767	(0.29850746266429046, 0.0)	(0.796334007245739, 0.8766816143301193, 0.7294776119266888)	0.657963441	0.887064
2	0.219101644	(0.36567164176375583, 0.0)	(0.797619042623378, 0.8516949152361929, 0.7499999999860074)	0.703427715	0.906561

I did not change anything in config.py except for the random seed, but the results are very low and obviously do not match with the paper. My environment is Python 3.6.5 and the PyTorch version is 1.0.1. Do you have any ideas about how to reproduce the results in your papers? Thanks so much for your help!