Closed bramiozo closed 2 years ago
I think this update solves this enough: https://github.com/umcu/negation-detection/commit/48366ce73bafb81cf113fdbb9e77ce7f1966a2df
It'll be difficult to reach truly equal splits per label if we want to take into account the number labels for negations & non-negations per file. Because we can't split labels from the same file, because that splits the text.
Optional check if file contains at least one categorical annotation.