损失函数:度量模型预测值与真实值之间的差异 对于回归问题:令残差r=f(X)−yr=f(X)−y L2损失:L2(r)=12r2L2(r)=12r2 L1损失:L1(r)=|r|L1(r)=|r| Huber损失:Lδ(r)={12r2,δ|r|−12δ2,if |r|<=δ if |r|>=δ Lδ(r)={12r2,if |r|<=δ δ|r|−12δ2,if |r|>=δ 对于分类问题: 0-1损失:l0/1(y,f(X))={10 yf(X)<0 otherewise l0/1(y,f(X))={1 yf(X)<0 0 otherewise Logistic损失:亦称负log似然损失/logloss llog(y,f(x))=log(1+exp(−yf(x)))llog(y,f(x))=log(1+exp(−yf(x))) 指数损失: 合页损失: 相关文章: 2022-02-07 2022-02-17 2022-12-23 2021-12-17 2022-12-23 2021-06-23