paper result（GNNer-AT model）
Opened this issue · 0 comments
xiaoding06 commented
Hello, The results of my GNNer-AT model run on scierc_datasets are wandb: epoch 39
wandb: macro 0.63747
wandb: micro 0.66351
wandb: overlap 987
wandb: prec 0.63492
wandb: rec 0.69479
wandb: train_loss 8.57301
wandb: trainer/global_step 3519
The recall value is 0.69, which is much higher than the 58.06+_0.86 in the paper. The macroF1 score is 0.63, which is 60.53+-0.69 in the paper, which is also higher. Also, the results seem to have no Num.Ov. value, only overlap 987. I would like to ask where to add the Num.ov. value in the code.