so i've been trying out bert_score to do some metrics, when testing it i'm finding that it's giving me some very high scores for some totally unrelated sentences. for example i tried this : refs = ["Ladies in a kitchen laughing at the camera while one grates cheese."] cands=["sky"] P,R,F = score(cands=cands,refs=refs,lang = 'en') print(P,R,F) the above execution returns : tensor([0.8362]) tensor([0.8028]) tensor([0.8192]) I'm not sure why it's giving me these but i would assume it's not what it should give me for those two sentences. Continue reading...
