SVAIGBA / WMSeg

MIT License
173 stars 42 forks source link

Can't reproduce the effect with no WM #10

Open Danielfile opened 3 years ago

Danielfile commented 3 years ago

I run your codes many times, but I can't reproduce the effect with no WM. For example, the (F, Roov) scores of PKU are (0.966448, 0.868110) ("BERT+SM+noWM") and (0.966335, 0.865612) ("BERT+CRF+noWM"), respectively. But the corresponding scores shown in your paper are (96.20, 84.43) and (96.32, 85.04), respectively. Why is this happening?

yuanheTian commented 3 years ago

Thanks for being interested in our paper.

We reported the experimental results of the baseline models when they are run on our machines. Due to machine variance, it is possible that you may get different results with the same hyper-parameter setting.