Hi Jonathan - just wondering how the hyperparameters for the slot filling baseline were chosen? The code doesn't do any early stopping based on a development split - were these based off some intuition about the dataset's relative sizes compared to the datasets which do have train/dev/test splits?
Hi Jonathan - just wondering how the hyperparameters for the slot filling baseline were chosen? The code doesn't do any early stopping based on a development split - were these based off some intuition about the dataset's relative sizes compared to the datasets which do have train/dev/test splits?
Thanks!