model - 回归验证中的 neg_mean_squared_error 和 mean_squared_error

Question

当我使用如下代码时，我完全感到困惑：

kfold = model_selection.KFold(n_splits=4, random_state=42, shuffle=True)

scoring1 = 'neg_mean_absolute_error'
scoring2 = 'r2'
scoring3 = 'neg_mean_squared_error'

results1 = model_selection.cross_val_score(lmodel, Xtrain, Ytrain, cv=kfold, scoring=scoring1)
results2 = model_selection.cross_val_score(lmodel, Xtrain, Ytrain, cv=kfold, scoring=scoring2)
results3 = model_selection.cross_val_score(lmodel, Xtrain, Ytrain, cv=kfold, scoring=scoring3)


print("MAE: %.10f (%.10f)" % (results1.mean(), results1.std()))

哪个值显示更好的性能？例如，当 score1 的数字分别为 -1.4 和 -2.5，model1 和 model2 和 model3 分别为 2.3 时，哪个模型在 score1 方面效果更好？

标签： modelregressioncross-validationk-foldmean-square-error