2021-11-03
作者:互联网
train_loss是训练集的损失值,test_loss是测试集的损失值
train_loss不断下降,test_loss不断下降,说明模型仍有训练空间;
train_loss不断下降,test_loss趋于不变,说明模型过拟合;
train_loss趋于不变,test_loss不断下降,说明数据集有问题;
train_loss趋于不变,test_loss趋于不变,说明需要减小学习率或调整步长;
标签:11,03,loss,趋于,说明,train,2021,test,不变 来源: https://blog.csdn.net/silly_miao/article/details/121128521