分布与切点
分布与切点
import numpy as np
pos = np.array([top_score(q) for q in answerable]) # 有答案
neg = np.array([top_score(q) for q in unanswerable]) # 无答案
for t in np.linspace(0, 1, 21):
fnr = (pos < t).mean() # 误拒
fpr = (neg >= t).mean() # 误收
print(round(t, 2), round(fnr, 3), round(fpr, 3))重排上线后阈值要重新校准：分数来源变了，旧阈值不再适用。