用LearningCurveDisplay.from_estimator在极度不平衡的数据集上画学习曲线，评估指标选average_precision，很容易撞上一条意思含糊的报错。换成 5 折或 10 折分层交叉验证也一样，于是怀疑是不是交叉验证策略没配对。问题现象复现代码：importnumpyasnpfromsklearn.ensembleimportRandomForestClassifierfromsklearn.model_selectionimportLearningCurveDisplay,train_test_splitfromimblearn