scoyer/fg2seq

Code for FG2SEQ: EFFECTIVELY ENCODING KNOWLEDGE FOR END-TO-END TASK-ORIENTED DIALOG.

fg2seq's Issues

May we have your best trained model's parameters or know the way of training it?

Hi, thanks so much for your pubic of code.

When I am trying to run the commands to train the model:
python myTrain.py -lr=0.001 -hdd=128 -dr=0.2 -bsz=16 -ds=kvr -B=10 -ss=10.0 or python myTrain.py -lr=0.001 -hdd=128 -dr=0.2 -bsz=8 -ds=cam -B=5 -ss=10.0, I found that the BLEU and F1 scores are not fixed each time re-train. And the automatic saved best model cannot achieve the same results you reported in the paper.

Therefore, I am wondering whether we can know how you trained the model and have access to your best model parameters.

Thanks so much!