Open syncdoth opened 3 years ago
Model architecture?
Maybe attention plots.
Data insight? How many percent of the words are which label, etc. -> if there are a lot of bio related labels and tokens, using biobert can be justified, etc.
Loss and accuracy plots. Also, other metrics such as F1 score might be a good idea.
Model architecture?
Maybe attention plots.
Data insight? How many percent of the words are which label, etc. -> if there are a lot of bio related labels and tokens, using biobert can be justified, etc.
Loss and accuracy plots. Also, other metrics such as F1 score might be a good idea.