Model weights sharing and training stopping criterion about paperrobot HOT 3 CLOSED

@zcwang0702 Hi, I just wait until the model stop. It will automatically select one with best dev score. The link prediction model usually run about 60-70 epochs. If you want to use the link prediction results for paper generation training, you can directly download the PubMed Term, Abstract, Conclusion, Title Dataset which already include the prediction results. Thanks!

from paperrobot.

@zcwang0702 Thank you very much for your interest in our work! Due to transition, I have lost all the trained model weight. However, the preferred trained parameters are already included in the code. I stop training based on the num of epochs and early stop patience. I didn't remember the exact epochs for training. However, the link prediction takes a week on the v100. Paper writing takes one day for training. Feel free to ask me any other questions.

from paperrobot.

from paperrobot.