Hey :)
Weird thing: as I actually had equivalent scores between CNN and RNN embeddings, my experiments with Optuna for now have yielded consistently lower result (by a big margin) on CNNs (I assume this change is in cause bca9516#diff-4853226f237bf1b0a17a302fbfa3e997 )
I cannot say if it's a regression before I actually retrain the same config as before but this might be something to look at in the next weeks.
Hey :)
Weird thing: as I actually had equivalent scores between CNN and RNN embeddings, my experiments with Optuna for now have yielded consistently lower result (by a big margin) on CNNs (I assume this change is in cause bca9516#diff-4853226f237bf1b0a17a302fbfa3e997 )
I cannot say if it's a regression before I actually retrain the same config as before but this might be something to look at in the next weeks.