Closed PolinaZulik closed 2 years ago
fine-tuned up to 2 epochs: Verb | P | R | F1 | Acc |
---|---|---|---|---|
Binary | 0.644 | 0.502 | 0.562 | 0.701 |
on train/val/test sample, 25 epochs, 1st epoch was the best in terms of val_loss and f1, so used the first epoch model. stopping on val loss. without quotes: 20 epochs, early stop with patience = 8. also first epoch model was the best. Desing | Verb | P | R | F1 | Acc |
---|---|---|---|---|---|
NLI with quotes | Binary | 0.60 | 0.71 | 0.65 | 0.71 |
NLI without quotes | Binary | 0.61 | 0.69 | 0.64 | 0.71 |
only examples with verb forms and lemmas present; only verbs for which both met and non-met examples are present: 4,740 examples. out of these 1,815 metaphoric (0.383) , 2,925 non-metaphoric (0.617). NLI, no fine-tuning, with
'cointegrated/rubert-base-cased-nli-twoway'
: