Closed ivanzvonkov closed 2 years ago
@hannah-rae FYI
Now this makes me wonder if decreasing dropout would help :D
A hparam search I did way back when suggested (some) dropout helped:
https://docs.google.com/spreadsheets/d/11HeMJo4mLnanEKHq8gH7DwBORazjg1uFBFPoy24TCJU/edit?usp=sharing
But this was not a formal investigation into the effects of dropout aha, just trying to get a sense of how the hyperparameters affected the model
I think I might still run it with 0.1 since there has been more data added
Result: Increasing dropout had a negative effect on the f1-score